Research map / Label-efficient and robust learning / Domain adaptation and generalization
Multimodal Cross-Domain Few-Shot Learning for Egocentric Action Recognition
Masashi Hatano, Ryo Hachiuma, Ryo Fujii, Hideo Saito
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 |
|---|---|---|---|
| add source cross-entropy classification loss to VideoMAE reconstruction pretraining (ablation) | MM-CDFSL · EPIC-Kitchens 5-shot | top-1 accuracy: 49.82 -> 58.70 | Improved |
| add target-domain reconstruction loss in pretraining (ablation) | MM-CDFSL · EPIC-Kitchens 5-shot | top-1 accuracy: 56.43 -> 58.70 | Improved |
| add multimodal distillation (optical flow and hand pose) to the RGB student (ablation) | MM-CDFSL · WEAR 5-shot | top-1 accuracy: 67.19 -> 69.57 | Improved |
| add optical flow distillation on top of RGB+Pose (ablation) | MM-CDFSL · WEAR 1-shot | top-1 accuracy: 49.39 -> 51.25 | Improved |
| set pretraining cross-entropy weight to 0.05 rather than 1 or 0.01 (ablation) | MM-CDFSL RGB only · EPIC-Kitchens 5-shot | top-1 accuracy: 54.58 / 52.40 -> 57.07 | Improved |
| apply ensemble masked inference (mask 0.75, P=2) instead of full-token inference | MM-CDFSL · Inference cost | runtime (ms): 22.1 -> 9.64 | Improved |
| use most accurate setting (no masking, P=1) versus CDFSL-V | MM-CDFSL · EPIC-Kitchens | accuracy: +5.23 points | Improved |
| use MM-CDFSL instead of Dynamic Distillation++ on MECCANO 5-shot | MECCANO · MECCANO 5-shot | top-1 accuracy: 37.87 -> 37.80 | No clear effect |
| use MM-CDFSL instead of CDFSL-V on WEAR 5-shot | WEAR · WEAR 5-shot | top-1 accuracy: 58.28 -> 69.57 | Improved |
Closest papers by meaning
- MMG-Ego4D: Multimodal Generalization in Egocentric Action Recognition (CVPR 2023 · similarity 0.89)
- Ego-Only: Egocentric Action Detection without Exocentric Transferring (ICCV 2023 · similarity 0.89)
- Interactive Prototype Learning for Egocentric Action Recognition (ICCV 2021 · similarity 0.89)
- Test-Time Adaptation for Combating Missing Modalities in Egocentric Videos (ICLR 2025 · similarity 0.88)
- X-MIC: Cross-Modal Instance Conditioning for Egocentric Action Generalization (CVPR 2024 · similarity 0.88)
- MMEgo: Towards Building Egocentric Multimodal LLMs for Video QA (ICLR 2025 · 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.
