Research map / Video / Action recognition
AWSD: Adaptive Weighted Spatiotemporal Distillation for Video Representation
Mohammad Tavakolian, Hamed R. Tavakoli, Abdenour Hadid
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 adaptive segment length selection (ablation) | WSD (ResNet-50) · UCF101 | accuracy: 86.4 -> 96.0 | Improved |
| add adaptive segment length selection (ablation) | WSD (ResNet-50) · HMDB51 | accuracy: 67.9 -> 75.6 | Improved |
| add adaptive segment length selection (ablation) | WSD (ResNet-50) · ActivityNet v1.3 | accuracy: 58.1 -> 90.4 | Improved |
| add adaptive segment length selection (ablation) | WSD (ResNet-50) · Maryland | accuracy: 73.6 -> 97.5 | Improved |
| use weighted-sum aggregation instead of average (ablation) | AWSD (ResNet-50) · UCF101 | accuracy: 91.5 -> 96.0 | Improved |
| use weighted-sum aggregation instead of max (ablation) | AWSD (ResNet-50) · ActivityNet v1.3 | accuracy: 85.6 -> 90.4 | Improved |
| increase fixed segment length (ablation) | WSD (ResNet-50) · ActivityNet v1.3 | accuracy: 57.8% to 28.1% | Hurt |
| apply AWSD to RGB frames instead of raw RGB frames | ResNet-50 · UCF101 | accuracy: 89.3 -> 96.0 | Improved |
| apply AWSD to optical flow instead of raw optical flow | ResNet-50 · UCF101 | accuracy: 90.1 -> 94.8 | Improved |
| use ResNet-101 instead of ResNet-50 backbone | AWSD · HMDB51 | accuracy: 75.6 -> 79.3 | Improved |
| use ResNet-50 instead of AlexNet backbone | AWSD · ActivityNet v1.3 | accuracy: 81.6 -> 90.4 | Improved |
| train on HMDB51 and test on UCF101 (cross-dataset) | AWSD (ResNet-50) · UCF101 | accuracy: 96.0 -> 88.6 | Hurt |
Closest papers by meaning
- DistInit: Learning Video Representations Without a Single Labeled Video (ICCV 2019 · similarity 0.85)
- Dancing with Still Images: Video Distillation via Static-Dynamic Disentanglement (CVPR 2024 · similarity 0.84)
- A Large-Scale Study on Unsupervised Spatiotemporal Representation Learning (CVPR 2021 · similarity 0.84)
- Motion Consistency Model: Accelerating Video Diffusion with Disentangled Motion-Appearance Distillation (NeurIPS 2024 · similarity 0.83)
- UDON: Universal Dynamic Online distillatioN for generic image representations (NeurIPS 2024 · similarity 0.83)
- Physics-Driven Spatiotemporal Modeling for AI-Generated Video Detection (NeurIPS 2025 · similarity 0.82)
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.
