Research map / Video / Action recognition
Mimetics: Towards Understanding Human Actions Out of Context
Philippe Weinzaepfel, Grégory Rogez
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 |
|---|---|---|---|
| mask humans in test videos (ablation) | 3D ResNeXt-101 trained on original Kinetics · Kinetics val | top-1 accuracy (%): 74.5 -> 38.7 | Hurt |
| train on videos with humans masked (ablation) | 3D ResNeXt-101 · Kinetics val (original test) | top-1 accuracy (%): 74.5 -> 65.7 | Hurt |
| use implicit pose features (SIP-Net) instead of explicit 3D pose | STGCN3D · Kinetics | mean accuracy (%): 10.6 -> 32.8 | Improved |
| use implicit pose features (SIP-Net) instead of explicit 3D pose | STGCN3D · HMDB51 | mean accuracy (%): 41.0 -> 51.2 | Improved |
| use pose-based SIP-Net instead of RGB 3D CNN on out-of-context actions | 3D-ResNeXt-101 RGB · Mimetics | top-1 accuracy (%): 8.6 -> 14.2 | Improved |
| use optical flow input instead of RGB | 3D-ResNeXt-101 · Mimetics | top-1 accuracy (%): 8.6 -> 11.8 | Improved |
| late-fuse RGB and Flow streams | 3D-ResNeXt-101 · Mimetics | top-1 accuracy (%): 11.8 -> 10.5 | Hurt |
| weakly-supervised pretraining on 65 million social-media videos | R(2+1)D-34 RGB · Mimetics | top-1 accuracy (%): 8.6 -> 11.3 | Improved |
| train SIP-Net only on the 50 Mimetics classes instead of 400 | SIP-Net · Mimetics | top-1 accuracy (%): 14.2 -> 25.1 | Improved |
| actor is a mime artist vs not | SIP-Net · Mimetics | top-1 accuracy (%): 17.8 -> 5.4 | Hurt |
| restrict to videos with no relevant object and scene | RGB 3D CNN · Mimetics | top-1 accuracy (%): 8.4 -> 4.5 | Hurt |
| use ground-truth tubes instead of estimated LCR tubes (ablation) | SIP-Net · JHMDB-1 | mean accuracy (%): 66.4 -> 73.3 | Improved |
Closest papers by meaning
- Quo Vadis, Skeleton Action Recognition? (IJCV 2021 · similarity 0.88)
- Pulling Actions out of Context: Explicit Separation for Effective Combination (CVPR 2018 · similarity 0.88)
- FineGym: A Hierarchical Video Dataset for Fine-Grained Action Understanding (CVPR 2020 · similarity 0.87)
- VideoNet: A Large-Scale Dataset for Domain-Specific Action Recognition (CVPR 2026 · similarity 0.87)
- ActionAtlas: A VideoQA Benchmark for Domain-specialized Action Recognition (NeurIPS 2024 · similarity 0.86)
- Disentangled Concepts Speak Louder Than Words: Explainable Video Action Recognition (NeurIPS 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.
