meta: title: Awesome Egocentric Atlas description: "Just launched Awesome Egocentric Atlas — a curated collection of first-person AI resources for egocentric vision, embodied AI, robotics, VLA, world models, WMA, memory, AR/VR, and hand-object interaction." last_major_audit: "2026-06-19" scope: Egocentric, first-person, wearable-camera, AR/VR headset, body/wrist camera, and ego-exo datasets, benchmarks, models, and tools. status_legend: open: Public download, public annotations, public code, or application-based access is clearly documented. request: Public project exists, but dataset access requires license, form, approval, or institutional agreement. benchmark: Mainly a benchmark, challenge, labels, or evaluation over existing videos. partial: Some assets are public, but full raw data, license, or long-term availability is unclear. watch: Recent or important resource whose release state still needs verification. resources: - name: Xperience-10M kind: dataset released: "2026" venue: "Hugging Face" year: 2026 status: request url: https://huggingface.co/datasets/ropedia-ai/xperience-10m scale: "10M experiences, 10K hours, six video streams, audio, stereo depth, camera pose, hand/body mocap, IMU, hierarchical language, about 1 PB total" modalities: [fisheye-video, stereo-video, audio, stereo-depth, camera-pose, hand-mocap, full-body-mocap, imu, hierarchical-language] tasks: [embodied-ai, world-modeling, robot-learning, sensor-fusion, 3d-4d-understanding, imitation-learning] created_by: Ropedia released_by: ropedia-ai on Hugging Face publisher: Hugging Face access: controlled non-commercial access; external agreement-signing step may be required license: other - name: Xperience-10M Sample kind: dataset released: "2026" venue: "Hugging Face" year: 2026 status: open url: https://huggingface.co/datasets/ropedia-ai/xperience-10m-sample scale: "Public sample episode; Hugging Face viewer reports 6 rows" modalities: [video, hdf5-annotations] tasks: [sample-data, loader-testing, visualization, task-suite-prototyping] created_by: Ropedia released_by: ropedia-ai on Hugging Face license: cc-by-nc-4.0 - name: Ego4D milestone: 2021 milestone_note: "The 3,670-hour massive-scale benchmark suite that catalyzed the modern egocentric era." kind: dataset released: "2021-10" venue: "CVPR 2022" year: 2021 status: request url: https://ego4d-data.org/ paper: https://arxiv.org/abs/2110.07058 scale: "3,670+ hours, 900+ camera wearers, 74 locations, 9 countries" modalities: [video, audio, gaze, stereo, 3d, narrations] tasks: [episodic-memory, nlq, moment-query, hand-object, forecasting, social, audio-visual] code: https://github.com/facebookresearch/Ego4d license: Ego4D License Agreement verified_at: "2026-06-15" - name: Ego-Exo4D milestone: 2023 milestone_note: "Synchronized ego and exo skilled-activity capture at scale; the reference for cross-view egocentric learning." kind: dataset released: "2023-11" venue: "CVPR 2024" year: 2023 status: request url: https://ego-exo4d-data.org/ paper: https://arxiv.org/abs/2311.18259 scale: "1,286 hours, 740 participants, synchronized ego/exo views" modalities: [video, audio, gaze, imu, 3d-point-clouds, camera-poses, language] tasks: [ego-exo, skilled-activity, cross-view, proficiency, pose] code: https://github.com/facebookresearch/projectaria_tools license: Ego-Exo4D License Agreement verified_at: "2026-06-15" - name: EPIC-KITCHENS-100 milestone: 2020 milestone_note: "The defining large-scale egocentric action-recognition benchmark and annual challenge suite." kind: dataset released: "2020-06" venue: "IJCV 2022" year: 2020 status: open url: https://epic-kitchens.github.io/ paper: https://arxiv.org/abs/2006.13256 scale: "100 hours, 20M frames, 90K action segments, 45 kitchens" modalities: [video, audio, narrations, object-boxes] tasks: [action-recognition, action-detection, anticipation, retrieval, domain-adaptation] code: https://github.com/epic-kitchens/epic-kitchens-100-annotations license: "CC BY-NC 4.0" license_url: https://creativecommons.org/licenses/by-nc/4.0/ verified_at: "2026-06-15" - name: HD-EPIC kind: dataset released: "2025-02" venue: "CVPR 2025" year: 2025 status: open url: https://arxiv.org/abs/2502.04144 paper: https://arxiv.org/abs/2502.04144 scale: "41 hours, 9 kitchens, 59K fine-grained actions, 26K VQA questions" modalities: [video, audio, gaze, object-masks, digital-twins, 3d] tasks: [vqa, action-recognition, audio-events, object-motion, gaze-reasoning] - name: HoloAssist kind: dataset released: "2023-09" venue: "ICCV 2023" year: 2023 status: open url: https://holoassist.github.io/ paper: https://arxiv.org/abs/2309.17024 scale: "166 hours, 350 instructor-performer pairs" modalities: [rgb, depth, gaze, hand, head, imu, audio] tasks: [interactive-assistance, mistake-detection, intervention-prediction, hand-forecasting] - name: Assembly101 kind: dataset released: "2022-03" venue: "CVPR 2022" year: 2022 status: open url: https://assembly-101.github.io/ paper: https://arxiv.org/abs/2203.14712 scale: "4,321 videos, 8 static plus 4 egocentric views, 1M fine-grained action segments" modalities: [multiview-video, egocentric-video, 3d-hand-pose] tasks: [procedural-action, anticipation, temporal-segmentation, mistake-detection] - name: Ego-1K kind: dataset released: "2026-03" venue: "CVPR 2026" year: 2026 status: open url: https://huggingface.co/datasets/facebook/ego-1k paper: https://arxiv.org/abs/2603.13741 scale: "Nearly 1,000 synchronized multiview egocentric videos" modalities: [multiview-video, calibration] tasks: [dynamic-scene-understanding, novel-view-synthesis, 3d-reconstruction] - name: EgoObjects kind: dataset released: "2023-09" venue: "ICCV 2023" year: 2023 status: open url: https://github.com/facebookresearch/EgoObjects paper: https://arxiv.org/abs/2309.08816 scale: "9K+ videos, 650K annotations, 368 categories in pilot release" modalities: [video, object-boxes, instance-labels] tasks: [object-detection, instance-understanding, continual-learning] - name: EgoBlind kind: dataset released: "2025-03" venue: "NeurIPS 2025 D&B" year: 2025 status: watch url: https://arxiv.org/abs/2503.08221 paper: https://arxiv.org/abs/2503.08221 scale: "1,392 first-person videos from blind and visually impaired users; 5,311 questions" modalities: [first-person-video, qa] tasks: [assistive-videoqa, blind-assistance, egocentric-vqa] created_by: Junbin Xiao, Nanxin Huang, Hao Qiu, Zhulin Tao, Xun Yang, Richang Hong, Meng Wang, Angela Yao release_note: NeurIPS 2025 D&B paper; verify stable dataset download and terms before benchmark use - name: EgoDex kind: dataset released: "2025-05" venue: "ICLR 2026" year: 2025 status: watch url: https://arxiv.org/abs/2505.11709 paper: https://arxiv.org/abs/2505.11709 scale: "829 hours, 194 tabletop tasks, Apple Vision Pro capture" modalities: [egocentric-video, 3d-hand-pose, head-pose] tasks: [dexterous-manipulation, imitation-learning, vla] - name: EgoVerse kind: dataset released: "2026-04" venue: "arXiv" year: 2026 status: watch url: https://egoverse.ai/ paper: https://arxiv.org/abs/2604.07607 scale: "1,362 hours, 80K episodes, 1,965 tasks, 2,087 demonstrators" modalities: [egocentric-video, annotations, tooling] tasks: [robot-learning, vla, human-demonstrations] - name: OpenEgo kind: dataset released: "2025-09" venue: "arXiv" year: 2025 status: watch url: https://arxiv.org/abs/2509.05513 paper: https://arxiv.org/abs/2509.05513 scale: "1,107 hours across six public datasets, 290 manipulation tasks" modalities: [egocentric-video, hand-pose, action-primitives] tasks: [dexterous-manipulation, standardized-pretraining, vla] - name: Dobb-E / HoNY kind: dataset released: "2023-11" venue: "arXiv" year: 2023 status: partial url: https://dobb-e.com/ paper: https://arxiv.org/abs/2311.16098 code: https://github.com/notmahi/dobb-e scale: "Homes of New York dataset: 13 hours, 22 homes, 216 environments, 5,620 trajectories, 1.5M RGB-D frames collected with an iPhone-based Stick" modalities: [rgb, depth, robot-actions, wrist-camera-video] tasks: [robot-learning, imitation-learning, manipulation, data-collection] verified_at: "2026-06-17" release_note: "Public dataset, code, model, and hardware pages exist; dataset terms should be rechecked before benchmark use." citation_key: dobbe_hony_2023 - name: Universal Manipulation Interface / UMI kind: toolkit released: "2024-02" venue: "RSS 2024" year: 2024 status: open url: https://umi-gripper.github.io/ paper: https://arxiv.org/abs/2402.10329 code: https://github.com/real-stanford/universal_manipulation_interface scale: "Hand-held GoPro gripper interface and policy stack for in-the-wild robot teaching, including example and cup-arrangement data workflows" modalities: [wrist-camera-video, imu, robot-actions] tasks: [robot-learning, imitation-learning, cross-embodiment-transfer, bimanual-manipulation, data-collection] license: MIT license_url: https://github.com/real-stanford/universal_manipulation_interface/blob/main/LICENSE verified_at: "2026-06-17" citation_key: umi_2024 - name: FastUMI kind: dataset released: "2024-09" venue: "arXiv" year: 2024 status: watch url: https://arxiv.org/abs/2409.19499 paper: https://arxiv.org/abs/2409.19499 scale: "UMI redesign with a reported 10,000+ real-world demonstration trajectories across 22 everyday manipulation tasks" modalities: [wrist-camera-video, robot-actions] tasks: [robot-learning, imitation-learning, manipulation, data-collection] verified_at: "2026-06-17" release_note: "Paper reports an open-sourced dataset; verify stable repository, download path, and license." citation_key: fastumi_2024 - name: FastUMI-100K kind: dataset released: "2025-10" venue: "arXiv" year: 2025 status: open url: https://huggingface.co/datasets/IPEC-COMMUNITY/FastUMI_100k_lerobot paper: https://arxiv.org/abs/2510.08022 code: https://github.com/MrKeee/FastUMI-100K scale: "100K+ UMI-style demonstration trajectories across 54 household manipulation tasks, standardized in LeRobot v2.1 with multi-view wrist-mounted fisheye images" modalities: [wrist-camera-video, robot-actions, language-instructions] tasks: [robot-learning, imitation-learning, cross-embodiment-transfer, bimanual-manipulation, data-collection] license: apache-2.0 license_url: https://github.com/MrKeee/FastUMI-100K/blob/main/LICENSE verified_at: "2026-06-17" citation_key: fastumi_100k_2025 - name: MV-UMI kind: toolkit released: "2025-09" venue: "arXiv" year: 2025 status: watch url: https://arxiv.org/abs/2509.18757 paper: https://arxiv.org/abs/2509.18757 scale: "Multi-view Universal Manipulation Interface adding third-person context to the egocentric wrist camera, evaluated across three manipulation tasks" modalities: [wrist-camera-video, third-person-video, robot-actions] tasks: [robot-learning, imitation-learning, cross-embodiment-transfer, manipulation, data-collection] verified_at: "2026-06-17" release_note: "arXiv paper found; verify public code, data, and license before treating as a reusable package." citation_key: mv_umi_2025 - name: UMIGen kind: model released: "2025-11" venue: "arXiv" year: 2025 status: watch url: https://arxiv.org/abs/2511.09302 paper: https://arxiv.org/abs/2511.09302 scale: "Cloud-UMI handheld collection and visibility-aware point-cloud generation for egocentric 3D observations and cross-embodiment imitation learning" modalities: [wrist-camera-video, point-clouds, robot-actions] tasks: [robot-learning, imitation-learning, cross-embodiment-transfer, 3d-reconstruction, data-collection] verified_at: "2026-06-17" release_note: "arXiv paper found; verify public Cloud-UMI code, generated data, and license." citation_key: umigen_2025 - name: Hoi! kind: dataset released: "2025-12" venue: "arXiv" year: 2025 status: watch url: https://arxiv.org/abs/2512.04884 paper: https://arxiv.org/abs/2512.04884 scale: "3,048 force-grounded articulated-manipulation sequences over 381 objects and 38 environments, spanning human hand, wrist camera, UMI gripper, and Hoi gripper embodiments" modalities: [wrist-camera-video, robot-actions, contact, articulated-objects] tasks: [robot-learning, imitation-learning, cross-embodiment-transfer, contact-understanding, hand-object] verified_at: "2026-06-17" release_note: "arXiv paper found; verify stable dataset download path and license." citation_key: hoi_force_grounded_2025 - name: YUBI kind: dataset released: "2026-06" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2606.10244 paper: https://arxiv.org/abs/2606.10244 scale: "UMI-based bimanual dexterous manipulation stack with a reported 8,434 hours, 1.20M episodes, and 119 tasks" modalities: [wrist-camera-video, robot-actions, 6dof] tasks: [robot-learning, imitation-learning, cross-embodiment-transfer, bimanual-manipulation, dexterous-manipulation] verified_at: "2026-06-17" release_note: "Paper reports release of hardware, software, and dataset; verify stable public repository and license." citation_key: yubi_2026 - name: Open X-Embodiment / RT-X kind: dataset released: "2023-10" venue: "ICRA 2024" year: 2023 status: open url: https://robotics-transformer-x.github.io/ paper: https://arxiv.org/abs/2310.08864 mirror: https://huggingface.co/datasets/jxu124/OpenX-Embodiment scope: adjacent scale: "1M+ real robot trajectories, 22 robot embodiments, 60 pooled robot datasets (robot-mounted and wrist cameras, not human first-person)" modalities: [robot-camera-video, wrist-camera-video, language-instructions, robot-actions, rlds] tasks: [vla, robot-learning, cross-embodiment-transfer, imitation-learning, manipulation] created_by: Open X-Embodiment Collaboration released_by: Open X-Embodiment Collaboration; unofficial Hugging Face mirror by jxu124 license: cc-by-4.0 scope_note: Adjacent robot-learning dataset included for VLA and robot-camera policy learning; not a human wearable-camera dataset. - name: EgoLive kind: dataset released: "2026-04" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2604.23570 paper: https://arxiv.org/abs/2604.23570 scale: "Large-scale real-world task-oriented egocentric routines" modalities: [egocentric-video, multimodal-annotations] tasks: [robot-manipulation, real-world-routines] - name: World In Your Hands kind: dataset released: "2025-12" venue: "arXiv" year: 2025 status: watch url: https://arxiv.org/abs/2512.24310 paper: https://arxiv.org/abs/2512.24310 scale: "Reported 1,000+ hours of multimodal human-centric manipulation data" modalities: [egocentric-video, multiview, depth, hand-pose, wrist-trajectory] tasks: [dexterous-manipulation, vla, human-centric-policy-learning] - name: InterVLA kind: dataset released: "2025-08" venue: "ICCV 2025" year: 2025 status: watch url: https://arxiv.org/abs/2508.04681 paper: https://arxiv.org/abs/2508.04681 scale: "11.4 hours, 1.2M frames, 2 egocentric and 5 exocentric videos" modalities: [egocentric-video, exocentric-video, human-motion, object-motion, verbal-commands] tasks: [vla, human-object-human-interaction, motion-estimation, interaction-prediction] - name: EgoPAT3D / EgoPAT3Dv2 kind: dataset released: "2024-03" venue: "ICRA 2024" year: 2024 status: open url: https://arxiv.org/abs/2403.05046 paper: https://arxiv.org/abs/2403.05046 scale: "Original EgoPAT3D reports 1M+ RGB-D and IMU frames; v2 expands 3D action-target prediction" modalities: [rgb, depth, imu, 3d-target-labels] tasks: [3d-action-target-prediction, hri, anticipation, manipulation-safety] - name: HOT3D milestone: 2024 milestone_note: "Reference benchmark for 3D hand-object tracking from AR glasses (Project Aria and Quest 3)." kind: dataset released: "2024-06" venue: "CVPR 2025" year: 2024 status: open url: https://facebookresearch.github.io/hot3d/ paper: https://arxiv.org/abs/2406.09598 scale: "833+ minutes, 3.7M+ images, 19 subjects, 33 objects" modalities: [aria, quest3, gaze, point-clouds, 3d-hand-pose, object-pose, camera-pose] tasks: [3d-hand-object-tracking, ar-vr, pose] license: "CC BY-NC-SA 4.0" license_url: https://creativecommons.org/licenses/by-nc-sa/4.0/ verified_at: "2026-06-15" - name: HOI4D license: "CC BY-NC 4.0" license_url: https://creativecommons.org/licenses/by-nc/4.0/ kind: dataset released: "2022-03" venue: "CVPR 2022" year: 2022 status: open url: https://hoi4d.github.io/ paper: https://arxiv.org/abs/2203.01577 scale: "2.4M RGB-D egocentric frames, 4,000+ sequences, 800 objects" modalities: [rgbd, point-clouds, 3d-hand-pose, object-pose, segmentation] tasks: [4d-hoi, pose-tracking, action-segmentation] - name: H2O kind: dataset released: "2021-04" venue: "ICCV 2021" year: 2021 status: open url: https://arxiv.org/abs/2104.11181 paper: https://arxiv.org/abs/2104.11181 modalities: [multiview-rgbd, 3d-hand-pose, object-pose, camera-pose, object-meshes, scene-point-clouds] tasks: [first-person-interaction-recognition, two-hand-pose, hand-object-pose] - name: ARCTIC kind: dataset released: "2022-04" venue: "CVPR 2023" year: 2022 status: request url: https://arctic.is.tue.mpg.de/ paper: https://arxiv.org/abs/2204.13662 scale: "2.1M frames" modalities: [video, hand-meshes, object-meshes, contact, articulated-objects] tasks: [bimanual-manipulation, reconstruction, interaction-field-estimation] - name: EgoHands kind: dataset released: "2015" venue: "ICCV 2015" year: 2015 status: open url: http://vision.soic.indiana.edu/projects/egohands/ scale: "48 Google Glass videos, 4,800 annotated images" modalities: [rgb, hand-masks] tasks: [hand-detection, hand-segmentation] - name: FPHA kind: dataset released: "2017" venue: "CVPR 2018" year: 2017 status: open url: https://guiggh.github.io/publications/first-person-hands/ scale: "100K+ RGB-D frames, 45 action classes, 26 objects" modalities: [rgbd, 3d-hand-pose, object-pose] tasks: [hand-action-recognition, 3d-hand-pose] - name: EgoHOS kind: dataset released: "2022-08" venue: "ECCV 2022" year: 2022 status: open url: https://github.com/owenzlz/EgoHOS paper: https://arxiv.org/abs/2208.03826 scale: "11,243 egocentric images" modalities: [rgb, hand-object-contact-masks] tasks: [hand-object-segmentation, contact-understanding] - name: Ego2Hands kind: dataset released: "2020-11" venue: "arXiv" year: 2020 status: open url: https://arxiv.org/abs/2011.07252 paper: https://arxiv.org/abs/2011.07252 modalities: [rgb, synthetic-composited-masks] tasks: [two-hand-segmentation, hand-detection] - name: Ego2HandsPose kind: dataset released: "2022-06" venue: "arXiv" year: 2022 status: open url: https://arxiv.org/abs/2206.04927 paper: https://arxiv.org/abs/2206.04927 modalities: [rgb, 3d-hand-pose] tasks: [two-hand-3d-pose] - name: EgoBody kind: dataset released: "2022" venue: "ECCV 2022" year: 2022 status: open url: https://sanweiliti.github.io/egobody/egobody.html modalities: [hololens2, rgb, depth, gaze, head-pose, hand-pose, body-pose] tasks: [3d-human-pose, body-shape, motion] - name: GIMO kind: dataset released: "2022-04" venue: "ECCV 2022" year: 2022 status: open url: https://github.com/y-zheng18/GIMO paper: https://arxiv.org/abs/2204.09443 modalities: [egocentric-video, gaze, body-pose, scene-scans] tasks: [gaze-informed-motion-prediction, human-motion, scene-context] - name: EgoHumans kind: benchmark released: "2023-05" venue: "ICCV 2023" year: 2023 status: partial url: https://arxiv.org/abs/2305.16487 paper: https://arxiv.org/abs/2305.16487 scale: "125K+ egocentric images" modalities: [egocentric-video, multiview-video, 2d-pose, 3d-pose, mesh] tasks: [multi-human-tracking, 3d-pose, mesh-recovery] - name: UnrealEgo kind: dataset released: "2022-08" venue: "ECCV 2022" year: 2022 status: open url: https://4dqv.mpi-inf.mpg.de/UnrealEgo/ paper: https://arxiv.org/abs/2208.01633 modalities: [synthetic-stereo, 3d-human-pose] tasks: [egocentric-3d-pose] - name: xR-EgoPose kind: dataset released: "2019" venue: "GitHub" year: 2019 status: open url: https://github.com/facebookresearch/xR-EgoPose modalities: [synthetic-egocentric, 3d-human-pose] tasks: [xr-pose-estimation] - name: EgoGTA / EgoPW-Scene kind: dataset released: "2022-12" venue: "arXiv" year: 2022 status: partial url: https://arxiv.org/abs/2212.11684 paper: https://arxiv.org/abs/2212.11684 modalities: [synthetic-egocentric, scene-depth, human-pose] tasks: [scene-aware-3d-human-pose, egocentric-depth, human-scene-interaction] - name: EgoTracks kind: benchmark released: "2023-01" venue: "NeurIPS 2023" year: 2023 status: open url: https://arxiv.org/abs/2301.03213 paper: https://arxiv.org/abs/2301.03213 derived_from: Ego4D tasks: [long-term-object-tracking, redetection] - name: TREK-150 kind: benchmark released: "2021" venue: "project page" year: 2021 status: open url: https://machinelearning.uniud.it/datasets/trek150/ derived_from: EPIC-KITCHENS tasks: [single-object-tracking] - name: Project Aria Datasets milestone: 2022 milestone_note: "Meta's research smart-glasses platform that opened the modern wave of AR and wearable egocentric data." kind: collection released: "2022" venue: "project page" status: open url: https://www.projectaria.com/datasets/ modalities: [aria, vrs, calibration, gaze, imu, mps] tasks: [ar-perception, scene-understanding, wearable-sensing] - name: Aria Digital Twin license: "CC BY-NC-SA 4.0" license_url: https://creativecommons.org/licenses/by-nc-sa/4.0/ kind: dataset released: "2023-06" venue: "arXiv" year: 2023 status: open url: https://www.projectaria.com/datasets/adt/ paper: https://arxiv.org/abs/2306.06362 scale: "200 sequences, 398 object instances" modalities: [aria, rgb, monochrome, imu, 6dof, depth, segmentation, synthetic-rendering] tasks: [3d-machine-perception, object-tracking, scene-reconstruction] - name: Aria Everyday Activities license: "CC BY-NC-SA 4.0" license_url: https://creativecommons.org/licenses/by-nc-sa/4.0/ kind: dataset released: "2024-02" venue: "arXiv" year: 2024 status: open url: https://www.projectaria.com/datasets/aea/ paper: https://arxiv.org/abs/2402.13349 scale: "143 daily activity sequences across five indoor locations" modalities: [aria, trajectory, point-cloud, gaze, speech] tasks: [everyday-activity, neural-scene-reconstruction, prompted-segmentation] - name: Nymeria license: "CC BY-NC-SA 4.0" license_url: https://creativecommons.org/licenses/by-nc-sa/4.0/ kind: dataset released: "2024-06" venue: "ECCV 2024" year: 2024 status: open url: https://www.projectaria.com/datasets/nymeria/ paper: https://arxiv.org/abs/2406.09905 scale: "300 hours, 264 participants, 50 locations" modalities: [aria, eye-tracking, imu, body-motion, language, observer-view] tasks: [motion-language, body-tracking, action-recognition] - name: Digital Twin Catalog kind: dataset released: "2025-04" venue: "CVPR 2025" year: 2025 status: open url: https://www.projectaria.com/datasets/dtc/ paper: https://arxiv.org/abs/2504.08541 scale: "2,000 scanned digital-twin objects" modalities: [3d-objects, dslr-images, aria-images] tasks: [object-reconstruction, digital-twin-evaluation] - name: EgoLife kind: dataset released: "2025-03" venue: "CVPR 2025" year: 2025 status: partial url: https://arxiv.org/abs/2503.03803 paper: https://arxiv.org/abs/2503.03803 scale: "300 hours, 6 participants, one week daily life" modalities: [meta-aria, video, audio, third-person-reference, qa] tasks: [life-assistant, long-context-qa, memory] - name: Ego-EXTRA kind: dataset released: "2025-12" venue: "arXiv" year: 2025 status: open url: https://fpv-iplab.github.io/Ego-EXTRA/ paper: https://arxiv.org/abs/2512.13238 scale: "50 hours, 15K+ VQA pairs" modalities: [video, dialogue, qa] tasks: [expert-trainee-assistance, egocentric-vqa] - name: EgoSchema milestone: 2023 milestone_note: "The benchmark that exposed how far models are from long-form egocentric video reasoning." kind: benchmark released: "2023-08" venue: "NeurIPS 2023" year: 2023 status: open url: http://egoschema.github.io/ paper: https://arxiv.org/abs/2308.09126 derived_from: Ego4D scale: "5K+ multiple-choice QA pairs, 250+ hours" tasks: [long-video-qa, temporal-reasoning] code: https://github.com/egoschema/EgoSchema verified_at: "2026-06-15" - name: X-LeBench kind: benchmark released: "2025-01" venue: "arXiv" year: 2025 status: watch url: https://arxiv.org/abs/2501.06835 paper: https://arxiv.org/abs/2501.06835 scale: "432 simulated life logs, 23 minutes to 16.4 hours" derived_from: Ego4D tasks: [extremely-long-video-understanding, lifelogging, long-context-qa] - name: TeleEgo kind: benchmark released: "2025-10" venue: "arXiv" year: 2025 status: watch url: https://arxiv.org/abs/2510.23981 paper: https://arxiv.org/abs/2510.23981 scale: "3,291 human-verified QA items in streaming setting" modalities: [egocentric-video, audio, text, timeline] tasks: [streaming-assistant, memory, real-time-understanding, cross-memory-reasoning] - name: EgoMemReason kind: benchmark released: "2026-05" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2605.09874 paper: https://arxiv.org/abs/2605.09874 derived_from: week-long egocentric video scale: "500 questions across entity, event, and behavior memory; average 25.9 hours of memory backtracking" tasks: [long-horizon-memory, entity-memory, event-memory, behavior-memory, egocentric-video-reasoning] created_by: Ziyang Wang, Yue Zhang, Shoubin Yu, Ce Zhang, Zengqi Zhao, Jaehong Yoon, Hyunji Lee, Gedas Bertasius, Mohit Bansal release_note: arXiv paper; verify public data/code release before benchmark use - name: EgoClip kind: dataset released: "2022-06" venue: "NeurIPS 2022" year: 2022 status: open url: https://github.com/showlab/EgoVLP paper: https://arxiv.org/abs/2206.01670 derived_from: Ego4D scale: "3.8M clip-text pairs" tasks: [video-language-pretraining, retrieval] - name: RefEgo kind: benchmark released: "2023-08" venue: "ICCV 2023" year: 2023 status: open url: https://github.com/shuheikurita/RefEgo paper: https://arxiv.org/abs/2308.12035 derived_from: Ego4D scale: "12K+ clips and 41 hours for video-based referring-expression comprehension" tasks: [referring-expression-comprehension, object-grounding, referred-object-tracking] created_by: Shuhei Kurita, Naoki Katsura, Eri Onami released_by: shuheikurita/RefEgo on GitHub - name: EgoBench kind: benchmark released: "2026-05" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2605.27820 paper: https://arxiv.org/abs/2605.27820 scale: "1,045 egocentric-video-grounded interactive tasks" tasks: [tool-using-agents, multimodal-reasoning, interaction] - name: EgoIntrospect kind: dataset released: "2026-05" venue: "arXiv" year: 2026 status: watch url: https://ego-introspect.github.io/ paper: https://arxiv.org/abs/2605.17262 scale: "180 hours, 60 subjects" modalities: [video, audio, gaze, motion, physiological-signals, self-annotations] tasks: [internal-state-reasoning, affect, intent, memory] - name: Minerva-Ego kind: benchmark released: "2025-05" venue: "arXiv" year: 2025 status: open scope: adjacent url: https://github.com/google-deepmind/neptune paper: https://arxiv.org/abs/2505.00681 scale: "Part of the Neptune / MINERVA long-video reasoning collection over web (YouTube) videos with manually annotated reasoning traces" tasks: [multistep-reasoning, long-video-qa, reasoning-traces] - name: MA-EgoQA kind: benchmark released: "2026-03" venue: "arXiv" year: 2026 status: open url: https://ma-egoqa.github.io/ paper: https://arxiv.org/abs/2603.09827 scale: "1.7K questions over multiple egocentric streams" tasks: [multi-agent-memory, social-reasoning, temporal-reasoning] - name: EASG-Bench kind: benchmark released: "2025-06" venue: "ICCV 2025 Workshop" year: 2025 status: open url: https://github.com/fpv-iplab/EASG-bench paper: https://arxiv.org/abs/2506.05787 tasks: [scene-graph-qa, relation-reasoning, temporal-reasoning] - name: MyEgo kind: benchmark released: "2026-04" venue: "CVPR 2026" year: 2026 status: open url: https://github.com/Ryougetsu3606/MyEgo paper: https://arxiv.org/abs/2604.01966 scale: "541 long videos, 5K personalized questions" tasks: [personalized-qa, ego-grounding, long-range-memory] - name: EgoToM kind: benchmark released: "2025-03" venue: "arXiv" year: 2025 status: watch url: https://arxiv.org/abs/2503.22152 paper: https://arxiv.org/abs/2503.22152 derived_from: Ego4D tasks: [theory-of-mind, goal-reasoning, belief-reasoning, next-action] created_by: Yuxuan Li, Vijay Veerabadran, Michael L. Iuzzolino, Brett D. Roads, Asli Celikyilmaz, Karl Ridgeway release_note: arXiv paper; verify public release route before benchmark use - name: EgoSound kind: benchmark released: "2026-02" venue: "CVPR 2026" year: 2026 status: watch url: https://arxiv.org/abs/2602.14122 paper: https://arxiv.org/abs/2602.14122 derived_from: [Ego4D, EgoBlind] scale: "7,315 validated QA pairs across 900 videos" tasks: [sound-understanding, audio-visual-reasoning, spatial-localization, causal-reasoning] created_by: Bingwen Zhu, Yuqian Fu, Qiaole Dong, Guolei Sun, Tianwen Qian, Yuzheng Wu, Danda Pani Paudel, Xiangyang Xue, Yanwei Fu release_note: CVPR 2026 paper; verify public release route before benchmark use - name: EgoEsportsQA kind: benchmark released: "2026-04" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2604.12320 paper: https://arxiv.org/abs/2604.12320 scale: "1,745 QA pairs from professional matches across three first-person shooter games" modalities: [first-person-video, qa] tasks: [first-person-qa, virtual-egocentric-reasoning, esports-perception, tactical-reasoning] created_by: Jianzhe Ma, Zhonghao Cao, Shangkui Chen, Yichen Xu, Wenxuan Wang, Qin Jin release_note: arXiv paper; virtual first-person setting, not wearable-camera capture - name: Causal-Plan-1M kind: benchmark released: "2026-06" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2606.01810 paper: https://arxiv.org/abs/2606.01810 scale: "Million-scale explicit reasoning traces over egocentric videos" tasks: [causal-reasoning, planning, next-state-estimation] - name: AssistQ kind: benchmark released: "2022-03" venue: "ECCV 2022" year: 2022 status: open url: https://showlab.github.io/assistq/ paper: https://arxiv.org/abs/2203.04203 scale: "531 QA samples from 100 instructional videos" tasks: [task-assistance, affordance-centric-qa, action-planning] - name: EgoTaskQA kind: benchmark released: "2022-10" venue: "NeurIPS 2022" year: 2022 status: benchmark url: https://arxiv.org/abs/2210.03929 paper: https://arxiv.org/abs/2210.03929 tasks: [task-understanding, causal-reasoning, counterfactual-qa, belief-reasoning] - name: HowToDIV kind: dataset released: "2025-08" venue: "arXiv" year: 2025 status: watch url: https://arxiv.org/abs/2508.11192 paper: https://arxiv.org/abs/2508.11192 scale: "507 conversations, 6,636 QA pairs, 24 hours" modalities: [egocentric-video, dialogue, qa] tasks: [procedural-task-assistance, instructional-dialogue, egocentric-qa] - name: EgoThink kind: benchmark released: "2023-11" venue: "CVPR 2024" year: 2023 status: benchmark url: https://arxiv.org/abs/2311.15596 paper: https://arxiv.org/abs/2311.15596 tasks: [first-person-vqa, vlm-evaluation, embodied-reasoning] - name: VidEgoThink kind: benchmark released: "2024-10" venue: "ICLR 2025" year: 2024 status: benchmark url: https://arxiv.org/abs/2410.11623 paper: https://arxiv.org/abs/2410.11623 derived_from: Ego4D tasks: [video-qa, hierarchy-planning, visual-grounding, reward-modeling] - name: EgoCoT-Bench kind: benchmark released: "2026-05" venue: "arXiv" year: 2026 status: open url: https://dstardust.github.io/EgoCoT/ paper: https://arxiv.org/abs/2605.19559 scale: "3,172 QA pairs over 351 egocentric videos" tasks: [grounded-cot, operation-centric-reasoning, evidence-consistency] - name: NoRA kind: benchmark released: "2026-06" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2606.04806 paper: https://arxiv.org/abs/2606.04806 scale: "1,420 first-person video clips" tasks: [normative-action-reasoning, grounded-reasonableness, safety] - name: Ego2Web kind: benchmark released: "2026-03" venue: "CVPR 2026" year: 2026 status: watch url: https://arxiv.org/abs/2603.22529 paper: https://arxiv.org/abs/2603.22529 tasks: [web-agents, egocentric-grounding, physical-digital-assistance] - name: Pause and Think kind: benchmark released: "2026-06" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2606.00616 paper: https://arxiv.org/abs/2606.00616 tasks: [assistive-action-suggestion, grounded-reasoning, temporal-consistency] - name: Ego-R1 kind: benchmark released: "2025-06" venue: "arXiv" year: 2025 status: watch url: https://arxiv.org/abs/2506.13654 paper: https://arxiv.org/abs/2506.13654 scale: "Ego-CoTT-25K, Ego-QA-4.4K, and Ego-R1 Bench" tasks: [ultra-long-video-qa, chain-of-tool-thought, temporal-retrieval] - name: EgoTrigger / HME-QA kind: benchmark released: "2025-08" venue: "ISMAR 2025 / TVCG" year: 2025 status: watch url: https://arxiv.org/abs/2508.01915 paper: https://arxiv.org/abs/2508.01915 derived_from: Ego4D scale: "340 human-annotated first-person QA pairs from full-length Ego4D videos curated for audio" modalities: [first-person-video, audio, qa] tasks: [memory-assistance, audio-triggered-capture, energy-efficient-smart-glasses, episodic-memory] created_by: Akshay Paruchuri, Sinan Hersek, Lavisha Aggarwal, Qiao Yang, Xin Liu, Achin Kulshrestha, Andrea Colaco, Henry Fuchs, Ishan Chatterjee release_note: Accepted to ISMAR 2025 as TVCG journal paper - name: GTEA / GTEA Gaze / EGTEA Gaze+ milestone: 2011 milestone_note: "Foundational hand-object and gaze egocentric activity datasets (GTEA, 2011) that seeded first-person action and attention research." kind: dataset released: "2018" venue: "project page" status: open url: https://cbs.ic.gatech.edu/fpv/ modalities: [video, gaze, action-labels] tasks: [action-recognition, gaze, hand-object] - name: Charades-Ego kind: dataset released: "2018-04" venue: "CVPR 2018" year: 2018 status: open url: https://prior.allenai.org/projects/charades paper: https://arxiv.org/abs/1804.09626 scale: "68.8 hours, 68,536 activity instances" modalities: [first-person-video, third-person-video, descriptions] tasks: [action-recognition, localization, captioning, domain-transfer] - name: MECCANO kind: dataset released: "2020-10" venue: "WACV 2021" year: 2020 status: open url: https://iplab.dmi.unict.it/MECCANO/ paper: https://arxiv.org/abs/2010.05654 modalities: [rgb, depth, gaze] tasks: [ehoi, active-object-detection, action-recognition, anticipation] - name: EgoProceL kind: dataset released: "2022-07" venue: "ECCV 2022" year: 2022 status: open url: https://sid2697.github.io/egoprocel/ paper: https://arxiv.org/abs/2207.10883 scale: "62 hours, 130 subjects, 16 tasks" tasks: [procedure-learning, key-step-localization] - name: EGO-CH kind: dataset released: "2020-02" venue: "arXiv" year: 2020 status: partial url: https://arxiv.org/abs/2002.00899 paper: https://arxiv.org/abs/2002.00899 scale: "27+ hours, 70 subjects, 2 cultural sites, 26 environments, 200+ points of interest" modalities: [egocentric-video, location-labels, poi-labels, surveys] tasks: [cultural-heritage, localization, object-recognition, retrieval, survey-prediction] - name: EgoExoLearn kind: dataset released: "2024-03" venue: "CVPR 2024" year: 2024 status: open url: https://github.com/OpenGVLab/EgoExoLearn paper: https://arxiv.org/abs/2403.16182 scale: "120 hours" modalities: [egocentric-video, demonstration-video, gaze, annotations] tasks: [cross-view-association, planning, skill-assessment] - name: DogCentric kind: dataset released: "2014" venue: "project page" status: open url: https://robotics.ait.kyushu-u.ac.jp/dog-centric-activity-dataset/ modalities: [animal-egocentric-video] tasks: [animal-activity-recognition] - name: KrishnaCam / OAK kind: dataset released: "2021-08" venue: "ICCV 2021" year: 2021 status: partial url: https://oakdata.github.io/ paper: https://arxiv.org/abs/2108.11005 scale: "OAK: 80 snippets, about 17.5 hours, 105 object categories from KrishnaCam" modalities: [egocentric-video, object-boxes] tasks: [continual-learning, object-detection, lifelogging] - name: EgoK360 kind: dataset released: "2020-10" venue: "arXiv" year: 2020 status: partial url: https://egok360.github.io/ paper: https://arxiv.org/abs/2010.08055 modalities: [first-person-360-video, action-labels] tasks: [360-video, activity-recognition] - name: EgoTraj kind: dataset released: "2026-05" venue: "arXiv" year: 2026 status: open url: https://github.com/yehiahmad/EgoTraj paper: https://arxiv.org/abs/2605.19004 scale: "75 Meta Quest Pro navigation sequences" modalities: [rgb, head-pose, gaze, scene-labels] tasks: [trajectory-prediction, navigation, assistive-systems] - name: VISOR kind: benchmark released: "2022" venue: "NeurIPS 2022" year: 2022 status: open url: https://epic-kitchens.github.io/VISOR/ derived_from: EPIC-KITCHENS license: cc-by-nc-4.0 verified_at: "2026-06-15" tasks: [hand-segmentation, active-object-segmentation, relations] - name: EPIC-Sounds kind: benchmark released: "2023" venue: "ICASSP 2023" status: open url: https://epic-kitchens.github.io/epic-sounds/ derived_from: EPIC-KITCHENS tasks: [audio-event-recognition] - name: EPIC-Fields kind: benchmark released: "2023" venue: "NeurIPS 2023" status: open url: https://epic-kitchens.github.io/epic-fields/ derived_from: EPIC-KITCHENS tasks: [3d-fields, spatial-reasoning] - name: EgoVLP milestone: 2022 milestone_note: "First egocentric video-language pretraining (EgoClip, EgoNCE) and a basis for ego representation learning." kind: model released: "2022-06" venue: "NeurIPS 2022" year: 2022 status: open url: https://github.com/showlab/EgoVLP paper: https://arxiv.org/abs/2206.01670 tasks: [video-language-pretraining, retrieval, ego4d-transfer] - name: EgoVLPv2 kind: model released: "2023-07" venue: "ICCV 2023" year: 2023 status: open url: https://shramanpramanick.github.io/EgoVLPv2/ paper: https://arxiv.org/abs/2307.05463 tasks: [video-language-pretraining, cross-modal-fusion] - name: LaViLa kind: model released: "2022-12" venue: "CVPR 2023" year: 2022 status: open url: https://arxiv.org/abs/2212.04501 paper: https://arxiv.org/abs/2212.04501 tasks: [video-language-representation, narration-generation] - name: EgoNCE++ kind: model released: "2024-05" venue: "ICLR 2025" year: 2024 status: open url: https://github.com/xuboshen/EgoNCEpp paper: https://arxiv.org/abs/2405.17719 tasks: [open-vocabulary-hoi, video-language-pretraining] - name: EgoDTM kind: model released: "2025-03" venue: "arXiv" year: 2025 status: open url: https://github.com/xuboshen/EgoDTM paper: https://arxiv.org/abs/2503.15470 tasks: [3d-aware-vlp, depth-text-pretraining] - name: EgoVLM kind: model released: "2025-06" venue: "arXiv" year: 2025 status: watch url: https://arxiv.org/abs/2506.03097 paper: https://arxiv.org/abs/2506.03097 tasks: [egocentric-video-reasoning, policy-optimization] - name: EgoGraph kind: model released: "2026-02" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2602.23709 paper: https://arxiv.org/abs/2602.23709 tasks: [temporal-knowledge-graph, ultra-long-video-qa] - name: Ropedia Xperience-10M Task Suite kind: benchmark released: "2026" venue: "Hugging Face" year: 2026 status: open url: https://huggingface.co/spaces/cy0307/ropedia-xperience-10m-task-suite derived_from: Xperience-10M Sample scale: "12 embodied-AI task contracts over one public Xperience-10M sample episode" modalities: [video, audio, depth, camera-pose, hand-mocap, full-body-mocap, imu, language] tasks: [task-design, sample-evaluation, multimodal-baselines, embodied-ai] created_by: cy0307 released_by: cy0307 on Hugging Face Spaces access: public Space and related artifacts - name: Ropedia Xperience-10M Task Baselines kind: model released: "2026" venue: "Hugging Face" year: 2026 status: open url: https://huggingface.co/cy0307/ropedia-xperience-10m-task-baselines derived_from: [Xperience-10M, Xperience-10M Sample] scale: "Public task definitions, baseline artifacts, metrics, and scale-up notes for Xperience-10M" modalities: [video, audio, depth, camera-pose, mocap, imu, language] tasks: [baseline-evaluation, task-suite, multimodal-representation, qwen3-omni, cosmos] created_by: cy0307 released_by: cy0307 on Hugging Face license: mit - name: GroundVQA kind: model released: "2023-12" venue: "CVPR 2024" year: 2023 status: open url: https://github.com/Becomebright/GroundVQA paper: https://arxiv.org/abs/2312.06505 tasks: [long-video-qa, temporal-grounding, ego4d-nlq] - name: HiERO kind: model released: "2025-05" venue: "arXiv" year: 2025 status: watch url: https://arxiv.org/abs/2505.12911 paper: https://arxiv.org/abs/2505.12911 tasks: [hierarchical-activity-reasoning, egomcq, egonlq, procedure-learning] - name: EgoAgent kind: model released: "2025-02" venue: "arXiv" year: 2025 status: watch url: https://arxiv.org/abs/2502.05857 paper: https://arxiv.org/abs/2502.05857 tasks: [joint-predictive-agent, future-state-prediction, action-prediction] - name: StillFast kind: model released: "2023-04" venue: "arXiv" year: 2023 status: open url: https://iplab.dmi.unict.it/stillfast/ paper: https://arxiv.org/abs/2304.03959 tasks: [short-term-object-interaction-anticipation, next-active-object, ego4d] - name: CONE kind: model released: "2022-11" venue: "ECCV 2022 Workshop" year: 2022 status: watch url: https://arxiv.org/abs/2211.08776 paper: https://arxiv.org/abs/2211.08776 tasks: [natural-language-query, video-language-grounding, ego4d-nlq] - name: EgoHandTrajPred / USST kind: model released: "2023-07" venue: "ICCV 2023" year: 2023 status: open url: https://actionlab-cv.github.io/EgoHandTrajPred paper: https://arxiv.org/abs/2307.08243 tasks: [3d-hand-trajectory-forecasting, anticipation, h2o, egopat3d] - name: Project Aria Tools kind: toolkit released: "2022" venue: "GitHub" status: open url: https://github.com/facebookresearch/projectaria_tools tasks: [vrs-loading, calibration, mps, gaze, trajectory] - name: EPIC-KITCHENS action models kind: toolkit released: "2019" venue: "GitHub" status: open url: https://github.com/epic-kitchens/action-models tasks: [action-recognition-baselines] - name: HOMIE-toolkit kind: toolkit released: "2026" venue: "GitHub" year: 2026 status: open url: https://github.com/Ropedia/HOMIE-toolkit derived_from: Xperience-10M tasks: [xperience-10m-loading, hdf5-annotations, visualization, rerun, calibration, depth, mocap, imu] created_by: Ropedia released_by: Ropedia on GitHub license: MIT # --- Catalog expansion (2026-06-14): README entries reconciled into the catalog --- - name: EgoCom / Ego audio-visual correspondence kind: dataset released: "2020" venue: "project page" status: open url: http://vision.cs.utexas.edu/projects/ego_av_corr/ scale: "Egocentric video with spatial audio for conversation and audio-visual correspondence tasks" tasks: [foundation-video, video-language] - name: EasyCom kind: dataset released: "2021-07" venue: "arXiv" year: 2021 status: open url: https://arxiv.org/abs/2107.04174 scale: "AR glasses egocentric multi-channel audio and wide-FOV RGB for noisy conversations" tasks: [foundation-video, video-language] - name: Look and Tell kind: dataset released: "2025-10" venue: "NeurIPS 2025 Workshop" year: 2025 status: watch url: https://arxiv.org/abs/2510.22672 scale: "25 participants, Project Aria plus stationary cameras, gaze/speech/video, 3D reconstructions" tasks: [foundation-video, video-language] - name: EgoVLA milestone: 2025 milestone_note: "Showed vision-language-action policies can be learned from egocentric human video and transferred to robots." kind: model released: "2025" venue: "project page" status: watch url: https://rchalyang.github.io/EgoVLA/ scale: "VLA training from egocentric human videos plus robot fine-tuning" tasks: [robot-learning, manipulation, vla] - name: EgoEngine kind: model released: "2026-06" venue: "arXiv" year: 2026 status: watch url: https://egoengine.github.io/ paper: https://arxiv.org/abs/2606.12604 scale: "Framework converting egocentric human manipulation videos into high-fidelity robot observation videos and executable robot action trajectories" tasks: [robot-learning, manipulation, vla] verified_at: "2026-06-16" - name: EgoAERO kind: model released: "2026-06" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2606.08057 scale: "Asset-free conversion from a single egocentric RGB-D demonstration; introduces EgoDex-R in the paper" tasks: [robot-learning, manipulation, vla] - name: HRDexDB kind: dataset released: "2026-04" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2604.14944 scale: "1.4K human/robot grasping trials, tactile, multiview video, egocentric video streams" tasks: [robot-learning, manipulation, vla] - name: UnrealEgo2 / UnrealEgo-RW kind: dataset released: "2024-01" venue: "arXiv" year: 2024 status: watch url: https://arxiv.org/abs/2401.00889 scale: "Expanded stereo egocentric pose datasets from synthetic and real-world capture" tasks: [hand-object, 3d, pose] - name: TouchMoment kind: dataset released: "2026-04" venue: "CVPR 2026 Findings" year: 2026 status: watch url: https://arxiv.org/abs/2604.12343 scale: "4,021 egocentric videos, 8,456 annotated hand-object contact moments" tasks: [hand-object, 3d, pose] - name: EgoFun3D kind: dataset released: "2026-04" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2604.11038 scale: "271 egocentric videos with 3D geometry, part segmentation, articulation and function-template annotations" tasks: [hand-object, 3d, pose] - name: EgoEMG kind: dataset released: "2026-05" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2605.05712 scale: "41 participants, bilateral EMG, IMU, RGB, external RGB-D, mocap hand labels" tasks: [hand-object, 3d, pose] - name: EgoEVHands kind: dataset released: "2024" venue: "GitHub" status: watch url: https://github.com/ZJUWang01/EgoEV-HandPose scale: "Stereo event-camera egocentric hand dataset with 5,419 sequences and 3D/2D keypoints" tasks: [hand-object, 3d, pose] - name: EventEgoHands kind: benchmark released: "2025-05" venue: "arXiv" year: 2025 status: watch url: https://arxiv.org/abs/2505.19169 scale: "Event-based egocentric 3D hand mesh reconstruction benchmark over N-HOT3D" tasks: [hand-object, 3d, pose] - name: A multimodal RGB/events FPV hand dataset kind: dataset released: "2026-06" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2606.10790 scale: "Synthetic event-based first-person hand detection from EgoHands plus v2e" tasks: [hand-object, 3d, pose] - name: EgoExoMem kind: dataset released: "2026-05" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2605.18734 scale: "2.6K MCQs across synchronized ego-exo videos" tasks: [memory, qa, assistant] - name: EgoSelf kind: model released: "2025" venue: "project page" status: watch url: https://abie-e.github.io/egoself_project/ scale: "Personalized egocentric assistant framework with graph memory" tasks: [memory, qa, assistant] - name: EgoCross kind: dataset released: "2025" venue: "GitHub" status: watch url: https://github.com/MyUniverse0726/EgoCross scale: "About 1,000 QA pairs over surgery, industry, extreme sports, and animal-perspective clips" tasks: [memory, qa, assistant] - name: ADL Dataset kind: dataset released: "2012" venue: "project page" status: partial url: https://www.csc.kth.se/cvap/actions/ scale: "Unscripted daily activity recordings with activity/object/hand annotations in classic literature" tasks: [action-recognition, procedure] - name: Wrist-mounted ADL kind: dataset released: "2015-11" venue: "CVPR 2016" year: 2015 status: open url: https://arxiv.org/abs/1511.06783 scale: "Synchronized head and wrist wearable-camera daily activities" tasks: [action-recognition, procedure] - name: Visual Experience Dataset / VEDB kind: dataset released: "2023" venue: "project page" status: partial url: http://tamaraberg.com/visualexperience/ scale: "240+ hours egocentric video with gaze/head tracking in classic literature" tasks: [action-recognition, procedure] - name: UT Ego kind: dataset released: "2012" venue: "project page" status: partial url: http://vision.cs.utexas.edu/projects/egocentric/ scale: "Long daily egocentric videos in classic summarization work" tasks: [action-recognition, procedure] - name: HUJI EgoSeg kind: dataset released: "2014" venue: "project page" status: partial url: https://www.vision.huji.ac.il/egoseg/ scale: "Long egocentric videos for temporal segmentation" tasks: [action-recognition, procedure] - name: JPL First-Person Interaction kind: dataset released: "2013" venue: "IEEE" status: partial url: https://ieeexplore.ieee.org/document/6909626 scale: "First-person videos of people interacting with a humanoid observer" tasks: [action-recognition, procedure] - name: FT-HID kind: dataset released: "2022" venue: "GitHub" status: open url: https://github.com/ENDLICHERE/FT-HID scale: "90K+ RGB-D first- and third-person human interaction samples from 109 subjects" tasks: [action-recognition, procedure] - name: LSC-ADL kind: dataset released: "2025-04" venue: "arXiv" year: 2025 status: open url: https://arxiv.org/abs/2504.02060 scale: "ADL annotations over lifelogging data generated with clustering plus human review" tasks: [action-recognition, procedure] - name: SEED4D kind: dataset released: "2024-12" venue: "WACV 2025" year: 2024 status: open scope: adjacent url: https://seed4d.github.io/ paper: https://arxiv.org/abs/2412.00730 scale: "Synthetic ego-exo dynamic 4D generator and autonomous-driving dataset (16.8M images, vehicle inward/outward cameras, LiDAR; WACV 2025)" tasks: [driving, 4d-reconstruction, multi-view] - name: Ego4D Benchmarks kind: benchmark released: "2021" venue: "project page" status: benchmark url: https://ego4d-data.org/ scale: "Derived benchmark suite built on Ego4D; tasks: Natural Language Query, Moment Query, episodic memory, state change, long-term anticipation, social/audio, hand-object" tasks: [benchmark, evaluation] - name: Ego-Exo4D Benchmarks kind: benchmark released: "2023" venue: "project page" status: benchmark url: https://ego-exo4d-data.org/ scale: "Derived benchmark suite built on Ego-Exo4D; tasks: Fine-grained activity, proficiency, cross-view translation, 3D pose, object correspondence" tasks: [benchmark, evaluation] - name: EPIC-KITCHENS Challenges kind: benchmark released: "2018" venue: "project page" status: benchmark url: https://epic-kitchens.github.io/ scale: "Derived benchmark suite built on EPIC-KITCHENS / EPIC-KITCHENS-100; tasks: Recognition, detection, anticipation, retrieval, domain adaptation" tasks: [benchmark, evaluation] - name: HD-EPIC VQA Challenge kind: benchmark released: "2025-02" venue: "arXiv" year: 2025 status: benchmark url: https://arxiv.org/abs/2502.04144 scale: "Derived benchmark suite built on HD-EPIC; tasks: Recipe, ingredient, nutrition, fine-grained action, 3D perception, object motion, gaze" tasks: [benchmark, evaluation] - name: EgoEnv kind: model released: "2022" venue: "project page" status: open url: https://vision.cs.utexas.edu/projects/ego-env/ scale: "Environment-aware representation learning from egocentric video" tasks: [video-language, representation-learning] - name: EgoMAS kind: model released: "2026" venue: "project page" status: open url: https://ma-egoqa.github.io/ scale: "Shared-memory baseline for multi-agent egocentric video QA" tasks: [video-language, representation-learning] - name: Ego-Exo representation transfer kind: model released: "2021-04" venue: "CVPR 2021" year: 2021 status: open url: https://arxiv.org/abs/2104.07905 scale: "Distillation from third-person video using ego-specific latent signals" tasks: [action, tracking, pose, hoi] - name: EgoPoseFormer kind: model released: "2024" venue: "GitHub" status: open url: https://github.com/ChenhongyiYang/egoposeformer scale: "Transformer baseline for stereo egocentric 3D human pose estimation" tasks: [action, tracking, pose, hoi] - name: EgoSTARK kind: model released: "2023-01" venue: "arXiv" year: 2023 status: open url: https://arxiv.org/abs/2301.03213 scale: "Adapted long-term tracker baseline for EgoTracks" tasks: [action, tracking, pose, hoi] - name: EgoHOS model kind: model released: "2022" venue: "GitHub" status: open url: https://github.com/owenzlz/EgoHOS scale: "Context-aware hand-object segmentation and augmentation pipeline" tasks: [action, tracking, pose, hoi] - name: AV-CONV kind: model released: "2023" venue: "project page" status: open url: https://vjwq.github.io/AV-CONV/ scale: "Audio-visual conversational graph prediction from ego/exo conversation" tasks: [action, tracking, pose, hoi] - name: EgoAction kind: model released: "2026-05" venue: "CVPR 2026" year: 2026 status: watch url: https://arxiv.org/abs/2605.24496 scale: "CVPR 2026 EPIC-KITCHENS action detection challenge pipeline" tasks: [action, tracking, pose, hoi] - name: EgoAdapt kind: model released: "2026-05" venue: "CVPR 2026" year: 2026 status: watch url: https://arxiv.org/abs/2605.24500 scale: "CVPR 2026 HD-EPIC VQA challenge inference-time adaptation pipeline" tasks: [action, tracking, pose, hoi] - name: Ego4D CLI and docs kind: toolkit released: "2021" venue: "project page" status: open url: https://ego4d-data.org/ scale: "Downloading and working with Ego4D data after license approval." tasks: [tooling] - name: Ego-Exo4D CLI and docs kind: toolkit released: "2023" venue: "project page" status: open url: https://ego-exo4d-data.org/ scale: "Downloading synchronized ego-exo data and annotations." tasks: [tooling] - name: VISOR API kind: toolkit released: "2022" venue: "GitHub" status: open url: https://github.com/epic-kitchens/VISOR scale: "Loading dense EPIC-KITCHENS hand/object masks and relations." tasks: [tooling] - name: EgoObjects API kind: toolkit released: "2023" venue: "GitHub" status: open url: https://github.com/facebookresearch/EgoObjects scale: "Working with category and instance-level egocentric object labels." tasks: [tooling] - name: HOT3D tooling kind: toolkit released: "2024" venue: "project page" status: open url: https://facebookresearch.github.io/hot3d/ scale: "Loading HOT3D hand/object/camera pose annotations and models." tasks: [tooling] - name: HOI4D tooling kind: toolkit released: "2022" venue: "project page" status: open url: https://hoi4d.github.io/ scale: "Loading RGB-D frames, point clouds, object meshes, and pose/segmentation annotations." tasks: [tooling] # --- Catalog expansion (2026-06-14): added from web research --- - name: EgoVid-5M kind: dataset released: "2024-11" venue: "NeurIPS 2025" year: 2024 status: open url: https://egovid.github.io/ paper: https://arxiv.org/abs/2411.08380 scale: "5M curated egocentric clips at 1080p with fine-grained kinematic and high-level textual action annotations (NeurIPS 2025)" tasks: [video-generation, world-modeling, action-conditioned-generation] - name: AoE kind: dataset released: "2026-02" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2602.23893 scale: "Always-on egocentric human video collection pipeline and corpus for embodied AI data scaling" tasks: [robot-learning, data-collection, embodied-ai] - name: Seeing Across Views (MV-RoboBench) kind: benchmark released: "2025-10" venue: "ICLR 2026" year: 2025 status: open scope: adjacent url: https://github.com/microsoft/MV-RoboBench paper: https://arxiv.org/abs/2510.19400 scale: "MV-RoboBench: 1.7K curated QA items over eight subtasks testing multi-view spatial reasoning of VLMs in robotic manipulation scenes (ICLR 2026)" tasks: [multi-view, robot-manipulation, vlm-evaluation] - name: EgoGesture kind: dataset released: "2017" venue: "IEEE TMM 2018" year: 2017 status: open url: https://ieeexplore.ieee.org/document/8299578/ scale: "24K+ gesture samples, 3M RGB-D frames, 50 subjects, 83 static and dynamic gestures across six indoor/outdoor scenes" tasks: [gesture-recognition, hand, wearable-interaction] - name: EgoBrain kind: dataset released: "2025-06" venue: "arXiv" year: 2025 status: watch url: https://arxiv.org/abs/2506.01353 scale: "Synchronized EEG and egocentric video for human action understanding from minds and eyes" tasks: [eeg, action-understanding, multimodal] - name: MM-Ego kind: model released: "2024-10" venue: "ICLR 2025" year: 2024 status: open url: https://arxiv.org/abs/2410.07177 scale: "Egocentric multimodal LLM with Memory Pointer Prompting; 7M-sample QA data engine and the EgoMemoria benchmark (629 videos, 7,026 questions)" tasks: [video-language, memory, qa] - name: EgoStream kind: benchmark released: "2026-05" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2605.31557 scale: "2,250 questions across seven memory dimensions with Answer Validity Windows, expanded to 8,528 recall-conditioned evaluations over streams up to 45.3 hours" tasks: [streaming-memory, episodic-memory, qa] - name: EgoMemory kind: benchmark released: "2026" venue: "OpenReview" year: 2026 status: watch url: https://openreview.net/forum?id=T0em4hJCQb scale: "165,795 user-specific object annotations over 245 videos from 45 participants for memory-augmented personalized retrieval" tasks: [personalized-retrieval, episodic-memory, long-context] - name: EgoTextVQA kind: benchmark released: "2025" venue: "CVPR 2025" year: 2025 status: open url: https://openaccess.thecvf.com/content/CVPR2025/papers/Zhou_EgoTextVQA_Towards_Egocentric_Scene-Text_Aware_Video_Question_Answering_CVPR_2025_paper.pdf scale: "Egocentric scene-text-aware video QA across indoor housekeeping and outdoor driving scenes (CVPR 2025)" tasks: [scene-text, qa, reading] - name: Gesture-Based Egocentric Video QA kind: benchmark released: "2026-03" venue: "CVPR 2026" year: 2026 status: watch url: https://arxiv.org/abs/2603.12533 scale: "Egocentric video QA grounded in the camera wearer's pointing and deictic gestures" tasks: [gesture-grounding, qa, referential] - name: EgoVQA kind: benchmark released: "2019" venue: "ICCV 2019" year: 2019 status: open url: https://openaccess.thecvf.com/content_ICCVW_2019/html/EPIC/Fan_EgoVQA_-_An_Egocentric_Video_Question_Answering_Benchmark_Dataset_ICCVW_2019_paper.html scale: "600+ QA pairs over egocentric videos; an early first-person video question answering benchmark" tasks: [qa, action, classic] - name: ExAct kind: benchmark released: "2025-06" venue: "arXiv" year: 2025 status: open url: https://arxiv.org/abs/2506.06277 scale: "Video-language benchmark for expert action analysis and feedback over skilled egocentric/exocentric activity" tasks: [expert-feedback, skill-assessment, video-language] - name: Home Action Genome / HOMAGE kind: dataset released: "2021-05" venue: "CVPR 2021" year: 2021 status: open url: https://homeactiongenome.org/ paper: https://arxiv.org/abs/2105.05226 scale: "27 participants, multi-modal synchronized ego and third-person views with 12 sensor types and hierarchical activity/action labels in home settings" tasks: [action-recognition, multi-view, compositional] - name: EgoExo-Fitness kind: dataset released: "2024-06" venue: "ECCV 2024" year: 2024 status: open url: https://github.com/iSEE-Laboratory/EgoExo-Fitness paper: https://arxiv.org/abs/2406.08877 scale: "Synchronized ego and exo fitness videos with keypoint verification, execution comments, and action quality scores (ECCV 2024)" tasks: [ego-exo, action-quality, skill-assessment] - name: WEAR kind: dataset released: "2023-04" venue: "IMWUT 2024" year: 2023 status: open url: https://mariusbock.github.io/wear/ paper: https://arxiv.org/abs/2304.05088 scale: "22 participants, 18 outdoor workout activities, synchronized egocentric video and 3D acceleration across 11 locations (IMWUT 2024)" tasks: [action-recognition, sensor-fusion, imu] - name: Egocentric HOI Detection kind: benchmark released: "2025-06" venue: "arXiv" year: 2025 status: open url: https://arxiv.org/abs/2506.14189 scale: "New benchmark and method for detecting hand-object interactions in egocentric video" tasks: [hand-object, detection, interaction] # --- Catalog expansion (2026-06-15): verified new resources --- - name: SHOW3D kind: dataset released: "2026-03" venue: "CVPR 2026" year: 2026 status: watch url: https://arxiv.org/abs/2603.28760 paper: https://arxiv.org/abs/2603.28760 scale: "In-the-wild 3D hand-object interaction dataset captured with a back-mounted multi-camera rig synchronized to a worn VR headset, with multi-view 3D shape/pose annotations and text descriptions" modalities: [multiview-video, 3d-hand-pose, object-pose, text] tasks: [hand-object, 3d-hand-object-tracking, 3d-reconstruction] raw_video_dependency: "self-captured in-the-wild video" scope_note: "Body-worn ego/ego-exo in-the-wild HOI capture (Meta Reality Labs, Yale)." verified_at: "2026-06-15" citation_key: show3d_2026 - name: EgoTouch / TouchAnything kind: dataset released: "2026-05" venue: "arXiv" year: 2026 status: watch url: https://jianyi2004.github.io/TouchAnything-Website/ paper: https://arxiv.org/abs/2605.13083 scale: "EgoTouch: 302 manipulation tasks across 4,530 episodes with synchronized egocentric and dual wrist-camera video, bimanual 3D hand pose (42 joints), and dense tactile pressure maps; TouchAnything is the accompanying tactile-estimation framework" modalities: [egocentric-video, wrist-camera-video, 3d-hand-pose, contact] tasks: [hand-object, contact-understanding, bimanual-manipulation] scope_note: "Egocentric plus wrist cameras with tactile pressure; emerging sensing." verified_at: "2026-06-15" citation_key: egotouch_touchanything_2026 - name: EgoEverything kind: benchmark released: "2026-04" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2604.08342 paper: https://arxiv.org/abs/2604.08342 scale: "5,000+ MCQ QA pairs over 100+ hours of egocentric video, with questions generated using human attention/gaze signals for AR long-context understanding" modalities: [egocentric-video, gaze, qa] tasks: [long-context, video-qa, gaze-reasoning] verified_at: "2026-06-15" citation_key: egoeverything_2026 - name: EgoNight kind: benchmark released: "2025-10" venue: "ICLR 2026" year: 2025 status: open url: https://insait-institute.github.io/EgoNight/ paper: https://arxiv.org/abs/2510.06218 code: https://github.com/dehezhang2/EgoNight scale: "Nighttime egocentric benchmark: EgoNight-VQA (3,658 QA over 90 videos, 12 QA types) plus day-night correspondence retrieval and egocentric depth estimation, with day-night aligned synthetic and real video" modalities: [egocentric-video, qa, depth, synthetic-rendering] tasks: [egocentric-vqa, egocentric-depth, retrieval] scope_note: "Low-light / nighttime robustness benchmark (INSAIT)." verified_at: "2026-06-15" citation_key: egonight_2025 - name: EgoIntent kind: benchmark released: "2026-03" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2603.12147 paper: https://arxiv.org/abs/2603.12147 scale: "3,014 steps across 15 indoor/outdoor daily-life scenarios, evaluating local intent (What), global intent (Why), and next-step plan (Next), with future-frame leakage removed" modalities: [egocentric-video, qa] tasks: [intent, next-action, anticipation] verified_at: "2026-06-15" citation_key: egointent_2026 - name: EgoMAGIC kind: dataset released: "2026-04" venue: "arXiv" year: 2026 status: open url: https://arxiv.org/abs/2604.22036 paper: https://arxiv.org/abs/2604.22036 data: https://doi.org/10.5281/zenodo.19239154 scale: "3,355 egocentric field-medicine videos over 50 medical tasks from a head-mounted stereo camera with audio; 1.95M labels for 124 medical objects and an action-detection challenge" modalities: [egocentric-video, stereo, audio, object-boxes] tasks: [action-detection, object-detection, procedural-task-assistance] raw_video_dependency: "self-captured field-medicine video" verified_at: "2026-06-15" citation_key: egomagic_2026 - name: Audio Hallucination in Egocentric Video kind: benchmark released: "2026-04" venue: "ICASSP 2026" year: 2026 status: watch url: https://arxiv.org/abs/2604.23860 paper: https://arxiv.org/abs/2604.23860 scale: "300 egocentric videos and 1,000 sound-focused questions probing audio hallucination in audio-visual LLMs that infer sounds from visual cues" modalities: [egocentric-video, audio, qa] tasks: [audio-visual-reasoning, sound-understanding, evaluation] verified_at: "2026-06-15" citation_key: audio_hallucination_ego_2026 - name: Ego2World kind: benchmark released: "2026-05" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2605.13335 paper: https://arxiv.org/abs/2605.13335 derived_from: HD-EPIC scale: "Compiles egocentric cooking videos (built on HD-EPIC) into executable symbolic worlds with graph-transition rules for belief-state planning under partial observability" modalities: [egocentric-video, annotations] tasks: [planning, belief-reasoning, embodied-reasoning] scope_note: "Egocentric-video-grounded executable simulator/benchmark, not a synthetic-only simulator." verified_at: "2026-06-15" citation_key: ego2world_2026 - name: EgoCrowds / CrowdEraser kind: dataset released: "2026-03" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2603.29036 paper: https://arxiv.org/abs/2603.29036 scale: "EgoCrowds: semi-synthetic paired crowded/empty clips from real egocentric walking-tour footage; CrowdEraser is a video-diffusion framework that removes crowds to produce humanless environment walkthroughs" modalities: [egocentric-video, first-person-video] tasks: [video-generation, scene-understanding, scene-reconstruction] scope_note: "First-person walking-tour video editing/generation." verified_at: "2026-06-15" citation_key: egocrowds_crowderaser_2026 # --- Catalog expansion (2026-06-15): growth additions --- - name: EgoPressure kind: dataset released: "2024-09" venue: "CVPR 2025" year: 2025 status: open url: https://yiming-zhao.github.io/EgoPressure/ paper: https://arxiv.org/abs/2409.02224 scale: "5.0 hours, 21 participants, a moving egocentric camera plus 7 stationary RGB-D cameras, with hand pose meshes and fine-grained per-contact touch pressure (CVPR 2025)" modalities: [egocentric-video, rgbd, 3d-hand-pose, contact] tasks: [hand-object, contact-understanding, 3d-hand-pose] verified_at: "2026-06-15" citation_key: egopressure_2025 - name: EgoTL kind: benchmark released: "2026-04" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2604.09535 paper: https://arxiv.org/abs/2604.09535 scale: "Think-aloud (say-before-act) egocentric capture with word-level spoken reasoning and metric-scale spatial annotations over 100+ household tasks; benchmarks VLMs and world models on six task dimensions" modalities: [egocentric-video, speech, annotations] tasks: [long-context, planning, embodied-reasoning] verified_at: "2026-06-15" citation_key: egotl_2026 - name: SuperMemory-VQA kind: benchmark released: "2026-06" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2606.00825 paper: https://arxiv.org/abs/2606.00825 scale: "52.9 hours of AI-glasses daily activities with RGB, audio transcription, gaze, IMU, and SLAM; 4,853 human-verified QA pairs across object/location memory, intent recall, and scene recall" modalities: [egocentric-video, audio, gaze, imu, point-clouds] tasks: [episodic-memory, long-context, qa] verified_at: "2026-06-15" citation_key: supermemory_vqa_2026 - name: EgoAVU kind: benchmark released: "2026-02" venue: "CVPR 2026" year: 2026 status: open url: https://github.com/facebookresearch/EgoAVU paper: https://arxiv.org/abs/2602.06139 code: https://github.com/facebookresearch/EgoAVU data: https://huggingface.co/datasets/facebook/EgoAVU_data project: https://cs20s030.github.io/EgoAVU/ scale: "Scalable audio-visual data engine producing EgoAVU-Instruct (3M QAs) and the human-verified EgoAVU-Bench (3K QAs) for egocentric audio-visual understanding (CVPR 2026 highlight)" modalities: [egocentric-video, audio, qa] tasks: [audio-visual-reasoning, sound-understanding, evaluation] verified_at: "2026-06-15" citation_key: egoavu_2026 - name: EgoScale milestone: 2026 milestone_note: "Revealed the log-linear data-scaling law for egocentric human-video VLA pretraining." kind: dataset released: "2026-02" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2602.16710 paper: https://arxiv.org/abs/2602.16710 scale: "20,854 hours of action-labeled egocentric human video for VLA pretraining with a human-to-robot two-stage transfer recipe; reveals a log-linear human-data scaling law" modalities: [egocentric-video, action-labels] tasks: [vla, robot-learning, dexterous-manipulation] scope_note: "Egocentric human-video corpus for dexterous-manipulation transfer." verified_at: "2026-06-15" citation_key: egoscale_2026 - name: StableHand kind: model released: "2026-05" venue: "arXiv" year: 2026 status: watch url: https://huajian-zeng.github.io/projects/stablehand/ paper: https://arxiv.org/abs/2605.18553 scale: "Quality-aware flow-matching baseline for world-space dual-hand motion estimation from egocentric video, evaluated on HOT3D and ARCTIC" modalities: [egocentric-video, 3d-hand-pose] tasks: [3d-hand-pose, pose, hand-object] verified_at: "2026-06-15" citation_key: stablehand_2026 # --- Catalog expansion (2026-06-16): egocentric models / methods --- - name: EgoVideo kind: model released: "2024-06" venue: "CVPR 2024" year: 2024 status: open url: https://github.com/OpenGVLab/EgoVideo paper: https://arxiv.org/abs/2406.18070 code: https://github.com/OpenGVLab/EgoVideo scale: "Egocentric video foundation model with a slow-fast adaptation recipe; winner of multiple Ego4D / EPIC-KITCHENS challenge tracks (CVPR 2024 EgoVis)" modalities: [egocentric-video, text] tasks: [foundation-video, video-language, representation-learning] verified_at: "2026-06-16" citation_key: egovideo_2024 - name: EgoHOD kind: model released: "2025-03" venue: "ICLR 2025" year: 2025 status: open url: https://github.com/InternRobotics/EgoHOD paper: https://arxiv.org/abs/2503.00986 code: https://github.com/InternRobotics/EgoHOD scale: "Fine-grained hand-object-dynamics pretraining (HOD pipeline) for egocentric video representation learning; SOTA on EK-100 retrieval/classification and EGTEA (ICLR 2025)" modalities: [egocentric-video, hand, text] tasks: [video-language-pretraining, hand-object, representation-learning] verified_at: "2026-06-16" citation_key: egohod_2025 - name: EgoM2P kind: model released: "2025-06" venue: "ICCV 2025" year: 2025 status: watch url: https://arxiv.org/abs/2506.07886 paper: https://arxiv.org/abs/2506.07886 scale: "Egocentric multimodal, multitask pretraining framework over RGB, depth, gaze, and camera pose with a masked modeling objective" modalities: [egocentric-video, depth, gaze, camera-pose] tasks: [multimodal, representation-learning, standardized-pretraining] verified_at: "2026-06-16" citation_key: egom2p_2025 - name: AlanaVLM kind: model released: "2024-06" venue: "arXiv" year: 2024 status: watch url: https://arxiv.org/abs/2406.13807 paper: https://arxiv.org/abs/2406.13807 scale: "Multimodal embodied-AI foundation model trained for egocentric video understanding with an egocentric video instruction dataset" modalities: [egocentric-video, text] tasks: [video-language, embodied-ai, video-qa] verified_at: "2026-06-16" citation_key: alanavlm_2024 - name: Vinci kind: model released: "2024-12" venue: "IMWUT 2025" year: 2025 status: open url: https://github.com/OpenGVLab/vinci paper: https://arxiv.org/abs/2412.21080 code: https://github.com/OpenGVLab/vinci scale: "Real-time, always-on embodied smart assistant built on an egocentric vision-language model for portable devices, with historical-context QA, task planning, and step-by-step visual demonstrations" modalities: [egocentric-video, audio, text] tasks: [real-time-understanding, streaming-assistant, task-assistance] verified_at: "2026-06-16" citation_key: vinci_2025 - name: AMEGO kind: model released: "2024-09" venue: "ECCV 2024" year: 2024 status: open url: https://gabrielegoletto.github.io/AMEGO/ paper: https://arxiv.org/abs/2409.10917 code: https://github.com/gabrielegoletto/AMEGO scale: "Active-memory representation from a single long egocentric video, capturing key locations and object interactions for fast multi-query answering; introduces the Active Memories Benchmark (20K+ queries over EPIC-KITCHENS, ECCV 2024)" modalities: [egocentric-video] tasks: [long-context, memory, episodic-memory] verified_at: "2026-06-16" citation_key: amego_2024 - name: EgoInstructor kind: model released: "2024-01" venue: "CVPR 2024" year: 2024 status: open url: https://jazzcharles.github.io/Egoinstructor/ paper: https://arxiv.org/abs/2401.00789 scale: "Retrieval-augmented egocentric video captioning that retrieves relevant third-person instructional clips, trained with an EgoExoNCE cross-view objective (CVPR 2024)" modalities: [egocentric-video, exocentric-video, text] tasks: [captioning, video-language, cross-view] verified_at: "2026-06-16" citation_key: egoinstructor_2024 - name: GroundNLQ kind: model released: "2023-06" venue: "CVPR 2023" year: 2023 status: open url: https://github.com/houzhijian/GroundNLQ paper: https://arxiv.org/abs/2306.15255 code: https://github.com/houzhijian/GroundNLQ scale: "Two-stage pretrained multi-scale grounding model for Ego4D Natural Language Queries; champion of the CVPR 2023 Ego4D NLQ challenge" modalities: [egocentric-video, text] tasks: [natural-language-query, temporal-grounding, ego4d] verified_at: "2026-06-16" citation_key: groundnlq_2023 - name: OSGNet kind: model released: "2025-05" venue: "CVPR 2025" year: 2025 status: open url: https://github.com/Yisen-Feng/OSGNet paper: https://arxiv.org/abs/2505.04270 code: https://github.com/Yisen-Feng/OSGNet scale: "Object-shot enhanced grounding network for egocentric video temporal grounding; top entry on Ego4D NLQ / Goal-Step moment queries (CVPR 2025)" modalities: [egocentric-video, text] tasks: [natural-language-query, temporal-grounding, object-grounding] verified_at: "2026-06-16" citation_key: osgnet_2025 # --- Catalog expansion (2026-06-16): 2026 H1 egocentric models --- - name: EgoThinker kind: model released: "2025-10" venue: "NeurIPS 2025" year: 2025 status: watch url: https://arxiv.org/abs/2510.23569 paper: https://arxiv.org/abs/2510.23569 scale: "Egocentric reasoning model with spatio-temporal chain-of-thought supervision and reinforcement fine-tuning for first-person video understanding" modalities: [egocentric-video, text] tasks: [egocentric-video-reasoning, grounded-cot, video-qa] verified_at: "2026-06-16" citation_key: egothinker_2025 - name: Walk through Paintings kind: model released: "2026-01" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2601.15284 paper: https://arxiv.org/abs/2601.15284 scale: "Egocentric world models built from internet priors, generating first-person walkthroughs of static scenes such as paintings" modalities: [egocentric-video] tasks: [world-modeling, video-generation, scene-understanding] verified_at: "2026-06-16" citation_key: walk_through_paintings_2026 - name: Temporal-Aware Ego VLM kind: model released: "2026-03" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2603.27184 paper: https://arxiv.org/abs/2603.27184 scale: "Training scheme that incentivizes temporal awareness in egocentric video-understanding models, improving order- and timing-sensitive reasoning" modalities: [egocentric-video, text] tasks: [temporal-reasoning, video-language, representation-learning] verified_at: "2026-06-16" citation_key: temporal_aware_ego_vlm_2026 - name: EgoMotion kind: model released: "2026-04" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2604.19105 paper: https://arxiv.org/abs/2604.19105 scale: "Hierarchical reasoning plus diffusion framework for egocentric vision-language motion generation from first-person context" modalities: [egocentric-video, motion, text] tasks: [motion, motion-estimation, video-language] verified_at: "2026-06-16" citation_key: egomotion_2026 - name: Gaze-SoM HOI Anticipation kind: model released: "2026-04" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2604.03667 paper: https://arxiv.org/abs/2604.03667 scale: "Leverages gaze and set-of-mark prompting in vision-language LLMs for hand-object-interaction anticipation from egocentric video" modalities: [egocentric-video, gaze, text] tasks: [anticipation, hand-object, gaze-reasoning] verified_at: "2026-06-16" citation_key: gaze_som_hoi_anticipation_2026 - name: Sanpo-D kind: benchmark released: "2026-01" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2601.18100 paper: https://arxiv.org/abs/2601.18100 scale: "Fine-grained re-annotation of the Sanpo egocentric navigation data with spatial signals, benchmarking spatial-conditioned reasoning of VLMs over long first-person videos" modalities: [egocentric-video, qa] tasks: [spatial-reasoning, navigation, video-qa] verified_at: "2026-06-16" citation_key: sanpo_d_2026 # --- Catalog expansion (2026-06-16): exhaustive scan additions --- - name: EgoSurgery kind: dataset year: 2024 released: "2024" venue: "MICCAI 2024" status: open url: https://github.com/Fujiry0/EgoSurgery paper: https://arxiv.org/abs/2503.18755 code: https://github.com/Fujiry0/EgoSurgery scale: "Egocentric open-surgery video dataset family: EgoSurgery-Phase for surgical phase recognition and EgoSurgery-HTS for pixel-wise hand-tool segmentation of 14 surgical tools (MICCAI 2024)" modalities: [egocentric-video, segmentation, object-boxes] tasks: [action-recognition, hand-object-segmentation, procedural-task-assistance] verified_at: "2026-06-16" citation_key: egosurgery_2024 - name: EgoEMS kind: dataset year: 2025 released: "2025-11" venue: "AAAI 2026" status: watch url: https://arxiv.org/abs/2511.09894 paper: https://arxiv.org/abs/2511.09894 scale: "High-fidelity multimodal egocentric dataset for cognitive assistance in emergency medical services, capturing time-critical team actions in high-stakes scenarios" modalities: [egocentric-video, audio, multimodal-annotations] tasks: [procedural-task-assistance, action-recognition, real-time-understanding] verified_at: "2026-06-16" citation_key: egoems_2025 - name: LEMMA kind: dataset year: 2020 released: "2020-07" venue: "ECCV 2020" status: open url: https://arxiv.org/abs/2007.15781 paper: https://arxiv.org/abs/2007.15781 scale: "Multi-view multi-agent multi-task daily-activity dataset across 14 kitchens and living rooms, densely annotated with atomic actions and human-object interactions (ECCV 2020)" modalities: [egocentric-video, third-person-video, multimodal-annotations] tasks: [compositional, action-recognition, anticipation] verified_at: "2026-06-16" citation_key: lemma_2020 - name: EgoMe kind: dataset year: 2025 released: "2025-01" venue: "arXiv" status: watch url: https://arxiv.org/abs/2501.19061 paper: https://arxiv.org/abs/2501.19061 scale: "Real-world dataset and challenge for following a demonstrator via egocentric view, pairing exocentric demonstrations with egocentric imitation across everyday tasks" modalities: [egocentric-video, exocentric-video] tasks: [imitation-learning, cross-view, action-understanding] verified_at: "2026-06-16" citation_key: egome_2025 - name: ENIGMA-360 kind: dataset year: 2026 released: "2026-03" venue: "arXiv" status: watch url: https://arxiv.org/abs/2603.09741 paper: https://arxiv.org/abs/2603.09741 scale: "Ego-exo dataset for human behavior understanding in industrial scenarios with 360-degree exocentric and egocentric capture" modalities: [egocentric-video, exocentric-video, first-person-360-video] tasks: [action-understanding, ego-exo, procedure] verified_at: "2026-06-16" citation_key: enigma_360_2026 - name: EgoIllusion kind: benchmark year: 2025 released: "2025-08" venue: "arXiv" status: watch url: https://arxiv.org/abs/2508.12687 paper: https://arxiv.org/abs/2508.12687 scale: "Benchmark for hallucinations in egocentric video understanding, probing where multimodal models fabricate objects, actions, and sounds" modalities: [egocentric-video, qa] tasks: [evaluation, multimodal-reasoning, audio-visual-reasoning] verified_at: "2026-06-16" citation_key: egoillusion_2025 - name: EgoExo-Con kind: benchmark year: 2025 released: "2025-10" venue: "arXiv" status: watch url: https://arxiv.org/abs/2510.26113 paper: https://arxiv.org/abs/2510.26113 scale: "Synchronized ego-exo video pairs with human-refined queries for view-invariant temporal verification and grounding; introduces the View-GRPO training framework" modalities: [egocentric-video, exocentric-video, qa] tasks: [cross-view, temporal-grounding, evaluation] verified_at: "2026-06-16" citation_key: egoexo_con_2025 - name: WearVQA kind: benchmark year: 2025 released: "2025-11" venue: "NeurIPS 2025" status: watch url: https://arxiv.org/abs/2511.22154 paper: https://arxiv.org/abs/2511.22154 scale: "First wearable-device VQA benchmark for smart glasses: 2,520 image-question-answer triplets across 7 image domains and 10 cognitive task types under occluded, low-light, and blurry egocentric capture (NeurIPS 2025)" modalities: [egocentric-video, qa] tasks: [vqa, evaluation, wearable-sensing] verified_at: "2026-06-16" citation_key: wearvqa_2025 - name: WearVox kind: benchmark year: 2026 released: "2026-01" venue: "arXiv" status: watch url: https://arxiv.org/abs/2601.02391 paper: https://arxiv.org/abs/2601.02391 scale: "Egocentric multichannel voice-assistant benchmark for wearables, evaluating spoken interaction grounded in first-person audio-visual context" modalities: [egocentric-video, audio, speech] tasks: [audio-visual-reasoning, streaming-assistant, evaluation] verified_at: "2026-06-16" citation_key: wearvox_2026 - name: SuperGlasses kind: benchmark year: 2026 released: "2026-02" venue: "CVPR 2026 Findings" status: watch url: https://arxiv.org/abs/2602.22683 paper: https://arxiv.org/abs/2602.22683 scale: "Benchmark evaluating vision-language models as intelligent agents for AI smart glasses across realistic wearable assistant tasks" modalities: [egocentric-video, qa] tasks: [tool-using-agents, evaluation, task-assistance] verified_at: "2026-06-16" citation_key: superglasses_2026 - name: PhysBrain kind: model year: 2025 released: "2025-12" venue: "arXiv" status: watch url: https://arxiv.org/abs/2512.16793 paper: https://arxiv.org/abs/2512.16793 scale: "Uses human egocentric data as a bridge to adapt vision-language models toward physical intelligence and embodied control" modalities: [egocentric-video, text] tasks: [embodied-ai, representation-learning, robot-learning] verified_at: "2026-06-16" citation_key: physbrain_2025 # --- Catalog expansion (2026-06-16): verified additions --- - name: EgoCampus kind: dataset released: "2025-12" venue: "arXiv" year: 2025 status: watch url: https://arxiv.org/abs/2512.07668 paper: https://arxiv.org/abs/2512.07668 scale: "Egocentric pedestrian eye-gaze dataset and model over campus walking routes with synchronized first-person video and gaze" modalities: [egocentric-video, gaze, scene-labels] tasks: [gaze, navigation, trajectory-prediction] verified_at: "2026-06-16" citation_key: egocampus_2025 - name: Egocentric Clinical Intent kind: benchmark released: "2026-01" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2601.06750 paper: https://arxiv.org/abs/2601.06750 scale: "Benchmark for egocentric clinical intent understanding by medical multimodal LLMs over first-person clinical procedure video" modalities: [egocentric-video, qa] tasks: [intent, procedural-task-assistance, evaluation] verified_at: "2026-06-16" citation_key: egocentric_clinical_intent_2026 # --- Catalog expansion (2026-06-16): late-2026 scan additions --- - name: 1M-HUGs / HUG kind: dataset released: "2026-06" venue: "arXiv" year: 2026 status: open url: https://grasping.io/ paper: https://arxiv.org/abs/2606.17054 data: https://grasping.io/ benchmark: https://grasping.io/ scale: "Egocentric human-grasp dataset with 1M frames (27.8 hours), 6,707 object instances across 41 buildings, plus HUG-Bench with 90 unseen objects for zero-shot grasping" modalities: [egocentric-video, rgbd, hand-pose, object-meshes] tasks: [robot-learning, dexterous-manipulation, hand-object, manipulation] verified_at: "2026-06-16" citation_key: hug_1m_2026 - name: EgoKit kind: toolkit released: "2026-05" venue: "arXiv" year: 2026 status: open url: https://egokit.chuange.org/ paper: https://arxiv.org/abs/2605.16797 scale: "Unified low-cost egocentric recording workflow across Android, iPhone, iPad, smart-glasses, and XR hosts, with synchronized ego and wrist views plus OpenXR hand tracking where available" modalities: [egocentric-video, wrist-camera-video, head-pose, hand-pose] tasks: [data-collection, robot-learning, wearable-sensing, tooling] verified_at: "2026-06-16" citation_key: egokit_2026 - name: MobileEgo Anywhere kind: toolkit released: "2026-05" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2605.05945 paper: https://arxiv.org/abs/2605.05945 scale: "Commodity-phone infrastructure for hour-plus egocentric trajectories; releases STERA processing, a mobile recording app, and a reported 200-hour / 584-session long-form dataset" modalities: [egocentric-video, trajectory, action-labels] tasks: [data-collection, robot-learning, vla, long-horizon-memory] verified_at: "2026-06-16" citation_key: mobileego_anywhere_2026 - name: UCS-Bench / DirectMe kind: benchmark released: "2026-06" venue: "ICML 2026" year: 2026 status: open url: https://github.com/cocowy1/UCS-Bench paper: https://arxiv.org/abs/2606.15200 code: https://github.com/cocowy1/UCS-Bench data: https://github.com/cocowy1/UCS-Bench scale: "170+ hours of egocentric observations with 8.1K+ timestamped questions for user-centric continual spatial intelligence; DirectMe maintains streaming spatial memory" modalities: [egocentric-video, qa, trajectory] tasks: [spatial-reasoning, episodic-memory, long-context, streaming-memory, qa] verified_at: "2026-06-16" citation_key: ucs_bench_directme_2026 - name: StreamMemBench kind: benchmark released: "2026-06" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2606.14571 paper: https://arxiv.org/abs/2606.14571 scale: "Streaming benchmark over EgoLife egocentric streams, pairing each evidence anchor with an initial task and a future follow-up to test recall, feedback incorporation, and reuse" modalities: [egocentric-video, qa, dialogue] tasks: [streaming-memory, memory-assistance, evaluation, egocentric-vqa] verified_at: "2026-06-16" citation_key: streammembench_2026 - name: Ego-METAS kind: benchmark released: "2026-05" venue: "arXiv" year: 2026 status: watch url: https://maria-sanvil.github.io/Ego-METAS-website/ paper: https://arxiv.org/abs/2606.02246 scale: "Online multimodal energy-efficient temporal action segmentation benchmark over 100+ hours from EgoExo4D, CMU-MMAC, and CaptainCook4D with RGB, audio, gaze, IMU, and monochrome streams" modalities: [egocentric-video, audio, gaze, imu, monochrome] tasks: [temporal-segmentation, action-segmentation, sensor-fusion, evaluation] verified_at: "2026-06-16" citation_key: ego_metas_2026 - name: Plan Watch Recover / EgoProactive kind: benchmark released: "2026-06" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2606.04970 paper: https://arxiv.org/abs/2606.04970 scale: "Proactive procedural-assistance benchmark introducing EgoProactive with out-of-plan annotations and Pro2Bench over Ego4D, EPIC-KITCHENS, EgoExo4D, HoloAssist, and HowTo100M" modalities: [egocentric-video, qa, action-labels] tasks: [procedural-task-assistance, mistake-detection, planning, interactive-assistance] verified_at: "2026-06-16" citation_key: plan_watch_recover_2026 - name: EgoExoBench kind: benchmark released: "2025-07" venue: "arXiv" year: 2025 status: watch url: https://arxiv.org/abs/2507.18342 paper: https://arxiv.org/abs/2507.18342 scale: "First-/third-person video reasoning benchmark with 7,300+ QA pairs across 11 subtasks covering semantic alignment, viewpoint association, and temporal reasoning" modalities: [egocentric-video, exocentric-video, qa] tasks: [ego-exo, cross-view, temporal-reasoning, video-qa, evaluation] verified_at: "2026-06-16" citation_key: egoexobench_2025 - name: EgoProx kind: benchmark released: "2026-05" venue: "CVPR 2026" year: 2026 status: watch url: https://arxiv.org/abs/2605.24456 paper: https://arxiv.org/abs/2605.24456 scale: "Egocentric 3D proximity reasoning benchmark organized across intention, exploration, exploitation, and chain-of-actions reasoning for MLLMs" modalities: [egocentric-video, qa, 3d] tasks: [spatial-reasoning, 3d, vqa, evaluation] verified_at: "2026-06-16" citation_key: egoprox_2026 - name: BARISTA kind: benchmark released: "2026-05" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2605.12074 paper: https://arxiv.org/abs/2605.12074 scale: "185 egocentric coffee-preparation videos with per-frame scene graphs, persistent object IDs, masks, tracks, boxes, attributes, relations, hand-object interactions, activities, and process steps" modalities: [egocentric-video, object-masks, object-boxes, action-labels] tasks: [scene-graph-qa, procedural-task-assistance, hand-object, temporal-reasoning] verified_at: "2026-06-16" citation_key: barista_2026 - name: TAVIS kind: benchmark released: "2026-05" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2605.07943 paper: https://arxiv.org/abs/2605.07943 scale: "Active-vision imitation-learning benchmark with TAVIS-Head (5 tasks), TAVIS-Hands (3 tasks), GR1T2 and Reachy2 embodiments, and the GALT anticipatory-gaze metric" modalities: [egocentric-video, gaze, robot-actions] tasks: [imitation-learning, gaze, robot-learning, evaluation] verified_at: "2026-06-16" citation_key: tavis_2026 - name: EgoPoint-Bench kind: benchmark released: "2026-04" venue: "ACL 2026" year: 2026 status: watch url: https://arxiv.org/abs/2604.21461 paper: https://arxiv.org/abs/2604.21461 scale: "11K+ simulated and real egocentric pointing samples spanning five evaluation dimensions and three levels of referential complexity" modalities: [egocentric-video, qa, hand-pose] tasks: [gesture-grounding, referential, spatial-reasoning, vqa] verified_at: "2026-06-16" citation_key: egopoint_bench_2026 - name: ReFocus / EM-QnF kind: benchmark released: "2026-04" venue: "CVPR 2026" year: 2026 status: watch url: https://nsubedi11.github.io/refocus paper: https://arxiv.org/abs/2604.24893 scale: "Episodic Memory with Questions and Feedback task for ambiguous egocentric NLQ, with feedback datasets and a plug-and-play Feedback Alignment Module" modalities: [egocentric-video, qa, dialogue] tasks: [episodic-memory, natural-language-query, interactive-assistance, temporal-grounding] verified_at: "2026-06-16" citation_key: refocus_em_qnf_2026 - name: HumanEgo kind: model released: "2026-05" venue: "arXiv" year: 2026 status: watch url: https://humanego-ai.github.io paper: https://arxiv.org/abs/2605.24934 scale: "Zero-shot robot learning from minutes of human egocentric videos via entity-level hand-object representations and flow-matching policies" modalities: [egocentric-video, hand-pose, robot-actions] tasks: [robot-learning, imitation-learning, cross-embodiment-transfer, hand-object] verified_at: "2026-06-16" citation_key: humanego_2026 - name: EgoGuide kind: model released: "2026-06" venue: "arXiv" year: 2026 status: watch url: https://silicx.github.io/EgoGuide paper: https://arxiv.org/abs/2606.14665 scale: "Robot-free demonstration collection with synchronized wrist and head/egocentric observations, online visual-geometric data-quality guidance, and a gated egocentric residual policy" modalities: [egocentric-video, wrist-camera-video, robot-actions] tasks: [data-collection, robot-learning, imitation-learning, vla] verified_at: "2026-06-16" citation_key: egoguide_2026 - name: Ego-Pi kind: model released: "2026-06" venue: "arXiv" year: 2026 status: watch url: https://egopipaper.github.io/ paper: https://arxiv.org/abs/2606.08107 scale: "VLA fine-tuning study over egocentric human and robot data using the pi0.5 foundation model and dexterous five-finger embodiments" modalities: [egocentric-video, robot-actions, hand-pose] tasks: [vla, robot-learning, cross-embodiment-transfer, dexterous-manipulation] verified_at: "2026-06-16" citation_key: ego_pi_2026 - name: EgoTactile kind: model released: "2026-06" venue: "ICML 2026 spotlight" year: 2026 status: watch url: https://egotactile.github.io/ paper: https://arxiv.org/abs/2606.09243 scale: "Benchmark and diffusion/baseline models pairing egocentric video with full-hand grasp-pressure supervision for everyday-object interactions" modalities: [egocentric-video, hand-pose, contact, depth] tasks: [contact-understanding, hand-object, robot-learning, evaluation] verified_at: "2026-06-16" citation_key: egotactile_2026 - name: EgoPressDiff kind: model released: "2026-06" venue: "ICASSP 2026" year: 2026 status: watch url: https://egopressdiff.github.io/ paper: https://arxiv.org/abs/2606.06872 scale: "Conditional video diffusion model for UV-domain egocentric hand-pressure maps, conditioned on hand pose, 3D mesh vertices, and depth" modalities: [egocentric-video, hand-pose, mesh, depth] tasks: [contact-understanding, hand-object, robot-learning] verified_at: "2026-06-16" citation_key: egopressdiff_2026 - name: EgoForce Hand Pose kind: model released: "2026-05" venue: "SIGGRAPH 2026" year: 2026 status: open url: https://dfki-av.github.io/EgoForce paper: https://arxiv.org/abs/2605.12498 code: https://dfki-av.github.io/EgoForce data: https://dfki-av.github.io/EgoForce scale: "Monocular egocentric 3D hand-pose and shape reconstruction across fisheye, perspective, and wide-FOV camera models, with source code, data, and demo available" modalities: [egocentric-video, hand-pose, hand-meshes] tasks: [3d-hand-pose, hand-object, ar-vr, pose] verified_at: "2026-06-16" citation_key: egoforce_hand_pose_2026 - name: PROSE kind: model released: "2026-06" venue: "arXiv" year: 2026 status: watch url: https://rckola.github.io/prose/ paper: https://arxiv.org/abs/2606.16569 scale: "Training-free RGB-only egocentric scene registration using VLM-derived object-level 3D scene graphs for persistent spatial memory in robots and AR" modalities: [egocentric-video, 3d, object-masks] tasks: [scene-reconstruction, spatial-reasoning, ar-perception, 3d] verified_at: "2026-06-16" citation_key: prose_2026 - name: OSGNet + MLLM Reranking kind: model released: "2026-05" venue: "CVPR 2026 EgoVis" year: 2026 status: open url: https://github.com/iLearn-Lab/CVPR25-OSGNet paper: https://arxiv.org/abs/2605.20818 code: https://github.com/iLearn-Lab/CVPR25-OSGNet scale: "Champion solution for the Ego4D Episodic Memory Challenge 2026 Natural Language Queries and GoalStep tracks, reranking OSGNet candidates with an MLLM" modalities: [egocentric-video, text] tasks: [natural-language-query, temporal-grounding, ego4d, video-language] verified_at: "2026-06-16" citation_key: osgnet_mllm_reranking_2026 - name: OmniEgo-R2 kind: model released: "2026-05" venue: "CVPR 2026 EgoVis" year: 2026 status: open url: https://github.com/Lee-zixu/OmniEgo-R2 paper: https://arxiv.org/abs/2605.24481 code: https://github.com/Lee-zixu/OmniEgo-R2 scale: "Routed reasoning framework for EgoCross; second place in both Source-Limited and Open-Source tracks of the 1st Cross-Domain EgoCross Challenge" modalities: [egocentric-video, qa] tasks: [video-qa, egocentric-video-reasoning, domain-transfer, grounded-reasoning] verified_at: "2026-06-16" citation_key: omniego_r2_2026 - name: CASTLE KG Retrieval kind: model released: "2026-06" venue: "CVPR 2026 EgoVis" year: 2026 status: watch url: https://arxiv.org/abs/2606.01933 paper: https://arxiv.org/abs/2606.01933 scale: "Third-place CASTLE 2026 challenge framework using video knowledge graphs and adaptive hierarchical retrieval over 600+ hours of synchronized 15-camera ego/exo streams" modalities: [egocentric-video, exocentric-video, multimodal-annotations] tasks: [long-context, multi-view, video-qa, temporal-reasoning] verified_at: "2026-06-16" citation_key: castle_kg_retrieval_2026 - name: Reflective Dialogue EgoCross kind: model released: "2026-05" venue: "CVPR 2026 EgoVis" year: 2026 status: watch url: https://arxiv.org/abs/2605.27885 paper: https://arxiv.org/abs/2605.27885 scale: "Inference-time Teacher/Solver reflective dialogue method for EgoCross, achieving third place in the Open-source Track without fine-tuning" modalities: [egocentric-video, qa, dialogue] tasks: [video-qa, egocentric-video-reasoning, domain-transfer, grounded-reasoning] verified_at: "2026-06-16" citation_key: reflective_dialogue_egocross_2026 - name: EARL kind: model released: "2026-05" venue: "ICML 2026" year: 2026 status: open url: https://github.com/yuggiehk/EARL paper: https://arxiv.org/abs/2605.14742 code: https://github.com/yuggiehk/EARL scale: "Analysis-guided RL framework for egocentric interaction reasoning and pixel grounding, with structured coarse-to-fine interaction parsing and GRPO optimization" modalities: [egocentric-video, object-masks, qa] tasks: [interaction, visual-grounding, segmentation, embodied-reasoning] verified_at: "2026-06-16" citation_key: earl_2026 - name: EgoExo-WM kind: model released: "2026-05" venue: "arXiv" year: 2026 status: watch url: https://vision.cs.utexas.edu/projects/EgoExo-WM/ paper: https://arxiv.org/abs/2605.15477 scale: "Transforms exocentric video into egocentric video using body-pose priors to train whole-body action-conditioned egocentric world models" modalities: [egocentric-video, exocentric-video, human-pose] tasks: [world-modeling, cross-view, action-conditioned-generation, human-motion] verified_at: "2026-06-16" citation_key: egoexo_wm_2026 - name: MotionGRPO kind: model released: "2026-05" venue: "ICML 2026" year: 2026 status: watch url: https://arxiv.org/abs/2605.05680 paper: https://arxiv.org/abs/2605.05680 scale: "GRPO-based reinforcement post-training for diffusion egocentric full-body 3D motion recovery from head-mounted device signals" modalities: [egocentric-video, body-pose, human-motion] tasks: [egocentric-3d-pose, motion-estimation, policy-optimization] verified_at: "2026-06-16" citation_key: motiongrpo_2026 - name: Hand Trajectory Fusion for Ego NLQ kind: model released: "2026-06" venue: "CVPR 2026 EgoVis" year: 2026 status: watch url: https://arxiv.org/abs/2606.02962 paper: https://arxiv.org/abs/2606.02962 scale: "Hand-trajectory encoder and cross-attention fusion for Ego4D Natural Language Query grounding, with gains on hand-object-interaction and quantity/state queries" modalities: [egocentric-video, hand-pose, text] tasks: [natural-language-query, temporal-grounding, hand-object, ego4d] verified_at: "2026-06-16" citation_key: hand_trajectory_fusion_ego_nlq_2026 - name: ACE-Ego-0 kind: model released: "2026-06" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2606.17200 paper: https://arxiv.org/abs/2606.17200 scale: "Unifies 4.53K hours of robot/sim data with 1.48K hours of pseudo-action-labeled egocentric human video via reliability-aware weighting for VLA pretraining; evaluated on RoboCasa GR1 and RoboTwin 2.0" modalities: [egocentric-video, language, robot-actions] tasks: [vla, robot-learning, imitation-learning, manipulation] release_note: "arXiv preprint (June 2026); watch for code and checkpoint release." verified_at: "2026-06-18" citation_key: ace_ego0_2026 - name: EgoXtreme kind: dataset released: "2026-03" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2603.25135 paper: https://arxiv.org/abs/2603.25135 scale: "Smart-glasses egocentric 6D object-pose dataset spanning industrial, sports, and rescue scenes with extreme motion blur, dynamic lighting, and heavy occlusion where current pose estimators fail" modalities: [egocentric-video, rgb] tasks: [3d-4d-understanding, pose-and-body] release_note: "arXiv preprint (March 2026); watch for dataset and benchmark release." verified_at: "2026-06-18" citation_key: egoxtreme_2026 - name: EgoForge kind: model released: "2026-03" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2603.20169 paper: https://arxiv.org/abs/2603.20169 scale: "Goal-directed egocentric world simulator that rolls out coherent first-person video from a single egocentric image and a high-level instruction" modalities: [egocentric-video, text] tasks: [world-modeling, video-generation, hand-object-interaction] release_note: "arXiv preprint (March 2026); watch for code and weights." verified_at: "2026-06-18" citation_key: egoforge_2026 - name: EgoHOI World Model kind: model released: "2026-03" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2603.13615 paper: https://arxiv.org/abs/2603.13615 scale: "Physics-informed egocentric world model that synthesizes photorealistic, contact-consistent hand-object interactions from action signals alone, without future-state inputs" modalities: [egocentric-video] tasks: [world-modeling, hand-object-interaction] release_note: "arXiv preprint (March 2026); watch for code release." verified_at: "2026-06-18" citation_key: egohoi_wm_2026 - name: EgoControl kind: model released: "2025-11" venue: "arXiv" year: 2025 status: watch url: https://arxiv.org/abs/2511.18173 paper: https://arxiv.org/abs/2511.18173 scale: "Pose-controllable egocentric video diffusion model that generates temporally coherent first-person frames conditioned on sequences of 3D full-body poses" modalities: [egocentric-video, body-pose] tasks: [video-generation, world-modeling, pose-and-body] release_note: "arXiv preprint (November 2025); watch for code release." verified_at: "2026-06-18" citation_key: egocontrol_2025 - name: EgoFlow kind: model released: "2026-04" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2604.01421 paper: https://arxiv.org/abs/2604.01421 scale: "Gradient-guided flow-matching framework that generates physically plausible 6DoF object trajectories from egocentric video using differentiable collision and smoothness constraints" modalities: [egocentric-video] tasks: [3d-4d-understanding, hand-object-interaction] release_note: "arXiv preprint (April 2026); watch for code release." verified_at: "2026-06-18" citation_key: egoflow_2026 - name: WholeBodyVLA kind: model released: "2025-12" venue: "ICLR 2026" year: 2025 status: open url: https://github.com/OpenDriveLab/WholebodyVLA paper: https://arxiv.org/abs/2512.11047 code: https://github.com/OpenDriveLab/WholebodyVLA scale: "Unified latent VLA that learns latent actions from action-free egocentric videos via a Latent Action Model for whole-body humanoid loco-manipulation; +21.3% over GR00T on AgiBot X2" modalities: [egocentric-video, language, robot-actions] tasks: [vla, robot-learning, manipulation] verified_at: "2026-06-19" citation_key: wholebodyvla_2025 - name: In-N-On kind: model released: "2025-11" venue: "arXiv" year: 2025 status: watch url: https://arxiv.org/abs/2511.15704 paper: https://arxiv.org/abs/2511.15704 scale: "Scales egocentric manipulation with 1,000+ hours of in-the-wild human egocentric video (PHSD) plus 20+ hours on-task data to train the Human0 language-conditioned flow-matching policy" modalities: [egocentric-video, language] tasks: [robot-learning, manipulation, imitation-learning] release_note: "arXiv preprint (November 2025); watch for code and data release." verified_at: "2026-06-19" citation_key: in_n_on_2025 - name: EgoZero kind: model released: "2025-05" venue: "arXiv" year: 2025 status: watch url: https://arxiv.org/abs/2505.20290 paper: https://arxiv.org/abs/2505.20290 scale: "Trains robot manipulation policies from Project Aria smart-glasses human demonstrations with zero robot training data; ~70% zero-shot success across 7 tasks on a Franka arm" modalities: [egocentric-video] tasks: [robot-learning, manipulation, imitation-learning] release_note: "arXiv preprint (May 2025); watch for code release." verified_at: "2026-06-19" citation_key: egozero_2025 - name: Ego-PM kind: model released: "2025-08" venue: "arXiv" year: 2025 status: watch url: https://arxiv.org/abs/2508.19852 paper: https://arxiv.org/abs/2508.19852 scale: "Egocentric predictive world model that jointly forecasts future actions and video frames conditioned on hand trajectories via a latent diffusion model; tested on Ego4D, BridgeData, and RLBench" modalities: [egocentric-video, hand-pose, text] tasks: [world-modeling, video-generation, robot-learning] release_note: "arXiv preprint (August 2025); watch for code release." verified_at: "2026-06-19" citation_key: ego_pm_2025 - name: PEDESTRIAN kind: dataset released: "2025-12" venue: "arXiv" year: 2025 status: watch url: https://arxiv.org/abs/2512.19190 paper: https://arxiv.org/abs/2512.19190 scale: "340 first-person pavement videos covering 29 urban sidewalk obstacle types, with deep-learning baselines for pedestrian-safety obstacle detection" modalities: [egocentric-video, rgb] tasks: [detection-segmentation, ar-sensing-navigation] release_note: "arXiv preprint (December 2025); watch for dataset release." verified_at: "2026-06-19" citation_key: pedestrian_2025 - name: DeltaDorsal kind: model released: "2026-01" venue: "arXiv" year: 2026 status: open url: https://github.com/hilab-open-source/deltadorsal paper: https://arxiv.org/abs/2601.15516 code: https://github.com/hilab-open-source/deltadorsal scale: "Dual-stream delta encoder that uses dorsal hand-skin deformation features to improve egocentric hand pose under self-occlusion (~18% lower mean per-joint angle error)" modalities: [egocentric-video, hand-pose] tasks: [pose-and-body, hand-object-interaction] verified_at: "2026-06-19" citation_key: deltadorsal_2026 - name: Edge Episodic Memory QA kind: model released: "2026-02" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2602.22455 paper: https://arxiv.org/abs/2602.22455 scale: "On-device dual-thread MLLM pipeline (continuous video-to-text summarization plus asynchronous querying) for real-time egocentric episodic-memory QA on QAEgo4D-Closed" modalities: [egocentric-video, text] tasks: [question-answering, memory-and-long-context] release_note: "arXiv preprint (February 2026); watch for code release." verified_at: "2026-06-19" citation_key: edge_episodic_memory_qa_2026 - name: LOME kind: model released: "2026-03" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2603.27449 paper: https://arxiv.org/abs/2603.27449 scale: "Action-conditioned egocentric world model that generates photorealistic human-object interactions from an image, text, and per-frame body/hand actions, with realistic physical consequences" modalities: [egocentric-video, body-pose, hand-pose, text] tasks: [world-modeling, video-generation, hand-object-interaction] release_note: "arXiv preprint (March 2026); watch for code release." verified_at: "2026-06-19" citation_key: lome_2026 - name: Ego-Nav Co-training kind: model released: "2026-06" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2606.01951 paper: https://arxiv.org/abs/2606.01951 scale: "Converts egocentric human walking videos into robot-action datasets and co-trains a VLA with robot demonstrations for mobile navigation (fruit-search task)" modalities: [egocentric-video, robot-actions, language] tasks: [vla, robot-learning, ar-sensing-navigation] release_note: "arXiv preprint (June 2026); watch for code release." verified_at: "2026-06-19" citation_key: ego_nav_cotraining_2026 - name: EXPLORE-Bench kind: benchmark released: "2026-03" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2603.09731 paper: https://arxiv.org/abs/2603.09731 scale: "Egocentric scene-prediction benchmark over real first-person videos: predict the final scene state after long atomic-action sequences, with object, attribute, and relation annotations; large human-model gap" modalities: [egocentric-video, text] tasks: [reasoning-intent-planning, question-answering, world-modeling] release_note: "arXiv preprint (March 2026); watch for benchmark release." verified_at: "2026-06-19" citation_key: explore_bench_2026 - name: Hand2World kind: model released: "2026-02" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2602.09600 paper: https://arxiv.org/abs/2602.09600 scale: "Autoregressive egocentric world model that generates photorealistic first-person interaction video from free-space hand gestures using occlusion-invariant hand conditioning and Plucker-ray camera embeddings; best on ARCTIC, HOT3D, HOI4D" modalities: [egocentric-video, hand-pose] tasks: [world-modeling, video-generation, hand-object-interaction] release_note: "arXiv preprint (February 2026); watch for code release." verified_at: "2026-06-19" citation_key: hand2world_2026 - name: EgoSim kind: model released: "2026-04" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2604.01001 paper: https://arxiv.org/abs/2604.01001 scale: "Closed-loop egocentric world simulator that generates spatially consistent first-person interaction videos with explicit 3D grounding and dynamic world-state updates for multi-stage simulation and cross-embodiment transfer" modalities: [egocentric-video, camera-pose] tasks: [world-modeling, video-generation, robot-learning] release_note: "arXiv preprint (April 2026); watch for code release." verified_at: "2026-06-19" citation_key: egosim_2026 - name: EgoPush kind: model released: "2026-02" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2602.18071 paper: https://arxiv.org/abs/2602.18071 scale: "End-to-end egocentric multi-object non-prehensile rearrangement for mobile robots from a single first-person camera, using object-centric relative-pose latents and teacher-student RL with zero-shot sim-to-real transfer" modalities: [egocentric-video] tasks: [robot-learning, manipulation, ar-sensing-navigation] release_note: "arXiv preprint (February 2026); watch for code release." verified_at: "2026-06-19" citation_key: egopush_2026 - name: MEgoHand kind: model released: "2025-05" venue: "arXiv" year: 2025 status: watch url: https://arxiv.org/abs/2505.16602 paper: https://arxiv.org/abs/2505.16602 scale: "Multimodal egocentric hand-object interaction motion generator pairing a VLM cerebrum with a flow-matching policy over a unified 3.35M RGB-D frame, 24K interaction, 1.2K object dataset" modalities: [egocentric-video, rgb-d, text, hand-pose] tasks: [hand-object-interaction, video-generation, pose-and-body] release_note: "arXiv preprint (May 2025); watch for code and data release." verified_at: "2026-06-19" citation_key: megohand_2025 - name: EgoH4 kind: model released: "2025-04" venue: "arXiv" year: 2025 status: watch url: https://arxiv.org/abs/2504.08654 paper: https://arxiv.org/abs/2504.08654 scale: "Diffusion transformer that forecasts both-hand 3D trajectories and poses from egocentric video and camera pose, including out-of-frame hands; trained on Ego-Exo4D (156K train / 34K eval sequences)" modalities: [egocentric-video, camera-pose, hand-pose] tasks: [pose-and-body, hand-object-interaction, world-modeling] release_note: "arXiv preprint (April 2025); watch for code release." verified_at: "2026-06-19" citation_key: egoh4_2025 - name: Diff-IP2D kind: model released: "2024-05" venue: "arXiv" year: 2024 status: open url: https://github.com/IRMVLab/Diff-IP2D paper: https://arxiv.org/abs/2405.04370 code: https://github.com/IRMVLab/Diff-IP2D scale: "Non-autoregressive diffusion model that jointly forecasts 2D hand trajectories and object affordances on egocentric video with explicit camera-egomotion conditioning" modalities: [egocentric-video, hand-pose] tasks: [hand-object-interaction, pose-and-body] verified_at: "2026-06-19" citation_key: diff_ip2d_2024 - name: IndEgo kind: dataset released: "2025-11" venue: "NeurIPS 2025" year: 2025 status: open url: https://huggingface.co/datasets/FraunhoferIPK/IndEgo paper: https://arxiv.org/abs/2511.19684 scale: "~197 hours of egocentric (plus ~97 hours exocentric) industrial collaborative-work video over assembly, logistics, inspection, and repair, with gaze, narration, sound, motion, hand pose, and point clouds" modalities: [egocentric-video, gaze, audio, motion, hand-pose, point-cloud] tasks: [action-and-procedure, assistance-and-agents, detection-segmentation] verified_at: "2026-06-19" citation_key: indego_2025 - name: LaMAria kind: benchmark released: "2025-09" venue: "arXiv" year: 2025 status: open url: https://lamaria.ethz.ch paper: https://arxiv.org/abs/2509.26639 scale: "City-scale egocentric visual-inertial SLAM benchmark from glasses-like wearables, hours/kilometers of trajectories with centimeter-accurate surveyed ground truth and difficulty-tiered tracks (incl. night and vehicle motion)" modalities: [egocentric-video, imu, camera-pose] tasks: [three-d-and-scene, ar-sensing-navigation, tracking] verified_at: "2026-06-19" citation_key: lamaria_2025 - name: EgoLifter kind: model released: "2024-03" venue: "ECCV 2024" year: 2024 status: open url: https://github.com/facebookresearch/egolifter paper: https://arxiv.org/abs/2403.18118 code: https://github.com/facebookresearch/egolifter scale: "Open-world 3D segmentation that decomposes natural egocentric video into individual 3D objects using 3D Gaussians and SAM weak supervision, with a transient module for dynamic objects (Aria Digital Twin benchmark)" modalities: [egocentric-video, camera-pose] tasks: [detection-segmentation, three-d-and-scene] verified_at: "2026-06-19" citation_key: egolifter_2024 - name: EgoSplat kind: model released: "2025-03" venue: "arXiv" year: 2025 status: watch url: https://arxiv.org/abs/2503.11345 paper: https://arxiv.org/abs/2503.11345 scale: "Open-vocabulary egocentric scene understanding with language-embedded 3D Gaussian splatting, using SAM2-tracked multi-view instance features and an instance-aware spatio-temporal module (gains on Aria Digital Twin)" modalities: [egocentric-video, language] tasks: [detection-segmentation, three-d-and-scene] release_note: "arXiv preprint (March 2025); watch for code release." verified_at: "2026-06-19" citation_key: egosplat_2025 - name: 3D-Aware Ego Instance Tracking kind: model released: "2024-08" venue: "arXiv" year: 2024 status: watch url: https://arxiv.org/abs/2408.09860 paper: https://arxiv.org/abs/2408.09860 scale: "3D-aware instance segmentation and multi-object tracking in egocentric video that lifts 2D masks with scene geometry to survive rapid motion and occlusion (large ID-switch reductions on EPIC Fields)" modalities: [egocentric-video, camera-pose] tasks: [detection-segmentation, tracking, three-d-and-scene] release_note: "arXiv preprint (August 2024); watch for code release." verified_at: "2026-06-19" citation_key: ego_3d_instance_tracking_2024 - name: EgoCogNav kind: model released: "2025-11" venue: "arXiv" year: 2025 status: watch url: https://arxiv.org/abs/2511.17581 paper: https://arxiv.org/abs/2511.17581 scale: "Cognition-aware egocentric navigation framework that forecasts human trajectory and head motion from perceived path uncertainty, modeling scanning, hesitation, and backtracking, with the 6-hour CEN dataset" modalities: [egocentric-video, motion] tasks: [ar-sensing-navigation, reasoning-intent-planning, pose-and-body] release_note: "arXiv preprint (November 2025); watch for dataset and code release." verified_at: "2026-06-19" citation_key: egocognav_2025 - name: Neck-Mounted Gaze (GLC) kind: model released: "2026-02" venue: "arXiv" year: 2026 status: watch url: https://arxiv.org/abs/2602.11669 paper: https://arxiv.org/abs/2602.11669 scale: "Transformer gaze estimator (GLC) for a shoulder-level neck-mounted camera with an out-of-bound gaze classification task and head/neck multi-view co-learning, plus a ~4-hour 8-subject dataset" modalities: [egocentric-video, gaze] tasks: [ar-sensing-navigation, pose-and-body] release_note: "arXiv preprint (February 2026); watch for code and data release." verified_at: "2026-06-19" citation_key: neck_gaze_glc_2026 - name: CMU-MMAC milestone: 2009 milestone_note: "The earliest egocentric dataset; launched first-person activity recognition at the first IEEE Workshop on Egocentric Vision (CVPR 2009)." kind: dataset released: "2009" venue: "CMU tech report 2009" year: 2009 status: open url: http://kitchen.cs.cmu.edu/ paper: http://kitchen.cs.cmu.edu/ scale: "Among the earliest egocentric datasets: a multimodal kitchen-activity database with head-mounted egocentric video plus body IMUs, motion capture, and audio for 43 subjects cooking 5 recipes" modalities: [egocentric-video, imu, motion, audio] tasks: [action-and-procedure, detection-segmentation] scope_note: "2009 CMU Multimodal Activity database; one of the first egocentric (ego+exo) activity datasets." verified_at: "2026-06-19" citation_key: cmu_mmac_2009 - name: First-Person Social Interactions kind: dataset released: "2012" venue: "CVPR 2012" year: 2012 status: open url: http://ai.stanford.edu/~alireza/publication/CVPR12.pdf paper: http://ai.stanford.edu/~alireza/publication/CVPR12.pdf scale: "The first egocentric social-interaction dataset: day-long head-mounted video of 8 subjects at a theme park, annotated for social interactions, roles, attention, and turn-taking" modalities: [egocentric-video] tasks: [audio-and-social, reasoning-intent-planning, detection-segmentation] verified_at: "2026-06-19" citation_key: fpsi_2012 - name: BEOID kind: dataset released: "2014" venue: "BMVC 2014" year: 2014 status: open url: https://dimadamen.github.io/BEOID/ paper: https://dimadamen.github.io/BEOID/ scale: "Gaze-tracked egocentric video of 8 users interacting with objects across 6 everyday locations (kitchen, workspace, printer, corridor, gym) for discovering task-relevant objects and interaction modes" modalities: [egocentric-video, gaze] tasks: [hand-object-interaction, action-and-procedure, detection-segmentation] verified_at: "2026-06-19" citation_key: beoid_2014