Research map / Video / Action recognition
Ego-Only: Egocentric Action Detection without Exocentric Transferring
Huiyu Wang, Mitesh Kumar Singh, Lorenzo Torresani
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 |
|---|---|---|---|
| replace K400 exocentric supervised pretraining with Ego4D MAE pretraining (ablation) | egocentric action detection · Ego4D MQ val | average mAP: 13.9 -> 16.3 | Improved |
| add MAE pretraining before temporal segmentation finetuning (ablation) | Ego-Only · Ego4D MQ val | average mAP: 4.2 -> 16.3 | Improved |
| add egocentric temporal segmentation finetuning to frozen MAE features (ablation) | Ego-Only · Ego4D MQ val | average mAP: 7.8 -> 16.3 | Improved |
| pretrain MAE on Ego4D instead of Kinetics-400 (ablation) | Ego-Only · Ego4D MQ val | average mAP: 13.4 -> 16.3 | Improved |
| add joint ego-exo MAE pretraining on K400 and Ego4D (ablation) | Ego-Only · Ego4D MQ val | average mAP: 16.3 -> 16.0 | No clear effect |
| add joint ego-exo finetuning with 515K labels (ablation) | Ego-Only · EPIC-Kitchens-100 detection | verb mAP: 29.0 -> 28.4 | No clear effect |
| scale exocentric pretraining from K400 to K600 (ablation) | exocentric transferring · EPIC-Kitchens-100 detection | verb mAP: 28.0 -> 27.1 | Hurt |
| increase MAE pretraining data from MQ clips to all Ego4D videos (ablation) | Ego-Only · Ego4D MQ val | average mAP: 14.5 -> 15.5 | Improved |
| use per-class reweighting instead of per-instance reweighting (ablation) | Ego-Only · Ego4D MQ val | average mAP: 16.3 -> 14.4 | Hurt |
| resample action centers instead of per-instance reweighting (ablation) | Ego-Only · Ego4D MQ val | average mAP: 16.3 -> 16.3 | No clear effect |
| scale backbone from ViT-B to ViT-L (ablation) | Ego-Only · Ego4D MQ val | average mAP: 16.3 -> 17.9 | Improved |
| concatenate features from last few blocks | frozen MAE features · Ego4D MQ val | average mAP | No clear effect |
Later papers that compared with this method
1 later accepted paper reports EgoOnly as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Streaming VideoLLMs for Real-Time Procedural Video Understanding (ICCV 2025 · better on 2 of 2 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Ego-Exo: Transferring Visual Representations From Third-Person to First-Person Videos (CVPR 2021 · similarity 0.92)
- EgoExo-Fitness: Towards Egocentric and Exocentric Full-Body Action Understanding (ECCV 2024 · similarity 0.91)
- Synchronization is All You Need: Exocentric-to-Egocentric Transfer for Temporal Action Segmentation with Unlabeled Synchronized Video Pairs (ECCV 2024 · similarity 0.90)
- Egocentric Action Recognition by Automatic Relation Modeling (TPAMI 2022 · similarity 0.90)
- Retrieval-Augmented Egocentric Video Captioning (CVPR 2024 · similarity 0.89)
- EgoVid-5M: A Large-Scale Video-Action Dataset for Egocentric Videos Generation (NeurIPS 2025 · similarity 0.88)
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.
