Research map / Video / Action recognition
Video Pose Distillation for Few-Shot, Fine-Grained Sports Action Recognition
James Hong, Matthew Fisher, Michaël Gharbi, Kayvon Fatahalian
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 RGB-only distillation to the VIPE* teacher features (ablation) | VIPE* · Diving48 (16-shot) | top-1 accuracy: 35.0 -> 54.9 | Improved |
| add RGB-only distillation to the VIPE* teacher features (ablation) | VIPE* · FX35 (16-shot) | top-1 accuracy: 75.7 -> 83.4 | Improved |
| add optical flow input without motion prediction (ablation) | VI-VPD distillation · Diving48 and FX35 (16-shot) | top-1 accuracy: 54.9 -> 53.7 (Diving48); 83.4 -> 83.2 (FX35) | Mixed |
| add motion prediction with auxiliary decoder D (ablation) | VI-VPD distillation · FX35 (16-shot) | top-1 accuracy: 83.2 -> 84.9 | Improved |
| add motion prediction with auxiliary decoder D (ablation) | VI-VPD distillation · Diving48 (16-shot) | top-1 accuracy: 53.7 -> 58.8 | Improved |
| add motion prediction (vs RGB-only distillation) (ablation) | VI-VPD distillation · FSJump6 (16-shot) | top-1 accuracy: 81.3 -> 80.2 | Mixed |
| distill on uncut video instead of action parts only (ablation) | VI-VPD · FX35 (16-shot) | top-1 accuracy: 84.3 -> 84.9 | Improved |
| also distill on test videos (no labels), entire corpus (ablation) | VI-VPD · Tennis7 (16-shot) | top-1 accuracy: 71.1 -> 72.6 | No clear effect |
| raise weak-pose score threshold from 0.1 to 0.5 (ablation) | VI-VPD · Diving48 (16-shot) | top-1 accuracy: 43.9 -> 58.8 | Improved |
| raise weak-pose score threshold from 0.5 to 0.9 (ablation) | VI-VPD · Diving48 (16-shot) | top-1 accuracy: 58.8 -> 25.7 | Hurt |
| add view invariance (VIPE* teacher instead of 2D joints) (ablation) | VPD · FX35 (16-shot) | top-1 accuracy: 82.7 -> 84.9 | Improved |
| crop around the athlete | GSM · FSJump6 (full training set) | top-1 accuracy: 42.1 -> 90.6 | Improved |
Closest papers by meaning
- SportsCap: Monocular 3D Human Motion Capture and Fine-Grained Understanding in Challenging Sports Videos (IJCV 2021 · similarity 0.87)
- FineSports: A Multi-person Hierarchical Sports Video Dataset for Fine-grained Action Understanding (CVPR 2024 · similarity 0.86)
- ViSTec: Video Modeling for Sports Technique Recognition and Tactical Analysis (AAAI 2024 · similarity 0.85)
- ActionAtlas: A VideoQA Benchmark for Domain-specialized Action Recognition (NeurIPS 2024 · similarity 0.85)
- BASKET: A Large-Scale Video Dataset for Fine-Grained Skill Estimation (CVPR 2025 · similarity 0.85)
- Seeing Beyond the Crop: Using Language Priors for Out-of-Bounding Box Keypoint Prediction (NeurIPS 2024 · similarity 0.85)
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.
