Research map / Video / Video-language understanding
Scaling Egocentric Vision: The EPIC-KITCHENS Dataset
Dima Damen, Hazel Doughty, Giovanni Maria Farinella, Sanja Fidler, Antonino Furnari, Evangelos Kazakos, Davide Moltisanti, Jonathan Munro, Toby Perrett, Will Price, Michael Wray
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| fuse spatial and temporal streams in TSN for action recognition | TSN · EPIC-KITCHENS action recognition S1 (seen kitchens) | top-1 action accuracy: 19.86 -> 20.54 | Improved |
| fuse spatial and temporal streams in TSN for action recognition | TSN · EPIC-KITCHENS action recognition S2 (unseen kitchens) | top-1 action accuracy: 10.11 -> 10.89 | Improved |
| use TSN (3 segments) instead of 2SCNN (1 segment) | action recognition baseline · EPIC-KITCHENS action recognition S1 (seen kitchens) | top-1 action accuracy: 13.23 -> 20.54 | Improved |
| test on unseen kitchens instead of seen kitchens | TSN (FUSION) · EPIC-KITCHENS action recognition | top-1 action accuracy: 20.5% -> 10.9% | Hurt |
| test on unseen kitchens instead of seen kitchens | Faster R-CNN · EPIC-KITCHENS object detection, all classes | mAP at IoU>0.5: 35.41 -> 33.11 | No clear effect |
| evaluate few-shot instead of many-shot classes | Faster R-CNN · EPIC-KITCHENS object detection S1 | mAP at IoU>0.5: 38.81 -> 20.72 | Hurt |
| raise IoU threshold from 0.5 to 0.75 | Faster R-CNN · EPIC-KITCHENS object detection S1, all classes | mAP: 35.41 -> 8.69 | Hurt |
| anticipate next action instead of recognising it | TSN (RGB) · EPIC-KITCHENS S1 | top-1 action accuracy: 19.86 -> 06.00 | Hurt |
| add flow stream and fusion to RGB TSN for anticipation | TSN (RGB) · EPIC-KITCHENS action anticipation S1 | top-1 action accuracy: 06.00 -> 04.62 | Hurt |
| use TSN instead of 2SCNN for anticipation | action anticipation baseline · EPIC-KITCHENS action anticipation S1 | top-1 action accuracy: 04.32 -> 06.00 | Improved |
| lengthen observation time tau_o beyond 1s | TSN anticipation · EPIC-KITCHENS action anticipation | accuracy | Hurt |
Closest papers by meaning
- The EPIC-KITCHENS Dataset: Collection, Challenges and Baselines (TPAMI 2020 · similarity 0.95)
- Rescaling Egocentric Vision: Collection, Pipeline and Challenges for EPIC-KITCHENS-100 (IJCV 2021 · similarity 0.93)
- HD-EPIC: A Highly-Detailed Egocentric Video Dataset (CVPR 2025 · similarity 0.92)
- EPFL-Smart-Kitchen: An Ego-Exo Multi-Modal Dataset for Challenging Action and Motion Understanding in Video-Language Models (NeurIPS 2025 · similarity 0.88)
- EgoVid-5M: A Large-Scale Video-Action Dataset for Egocentric Videos Generation (NeurIPS 2025 · similarity 0.88)
- Ego-Exo4D: Understanding Skilled Human Activity from First- and Third-Person Perspectives (CVPR 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
