Research map / Video / Action recognition
Rethinking Resolution in the Context of Efficient Video Recognition
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 clip-level KD from a high-resolution teacher (ablation) | ResNet-50 at 112p · ActivityNet-v1.3 | mAP: 71.8\% -> 73.4\% | Helped |
| switch from clip-level to frame-level KD (ablation) | ResKD · ActivityNet-v1.3 | mAP: 73.4\% -> 76.6\% | Helped |
| switch from frame-level to pixel-level (feature map) KD (ablation) | ResKD · ActivityNet-v1.3 | mAP: 76.6\% -> 78.5\% | Helped |
| use high-resolution teacher input instead of only a larger teacher at low resolution (ablation) | ResKD · ActivityNet-v1.3 | mAP: 73.3\% -> 76.2\% | Helped |
| use R152 teacher in addition to high-resolution input (ablation) | ResKD · ActivityNet-v1.3 | mAP: 76.2\% -> 78.5\% | Helped |
| apply ResKD to SlowOnly at 112p | SlowOnly-ResNet50 · Kinetics-400 | top-1 accuracy: 69.3\% -> 73.1\% | Helped |
| apply ResKD to Swin_S at 112p | Swin_S · Kinetics-400 | top-1 accuracy: 76.3\% -> 80.0\% | Helped |
| apply ResKD (low resolution) to TSM | TSM-ResNet50 · Something-Something V2 | top-1 accuracy: 59.1\% -> 60.6\% | Helped |
| reduce input resolution from 224p to 56p with ResKD (ablation) | ResKD ResNet-50 · ActivityNet-v1.3 | mAP: 81.7\% -> 70.5\% | Hurt |
| modify ResNet-50 strides from (1,2,2,2) to (1,1,2,2) (ablation) | ResNet-50 · ActivityNet-v1.3 | mAP at low resolution | Mixed |
Closest papers by meaning
- Recurrent Back-Projection Network for Video Super-Resolution (CVPR 2019 · similarity 0.86)
- Space-Time Distillation for Video Super-Resolution (CVPR 2021 · similarity 0.86)
- Multi-Scale Aligned Distillation for Low-Resolution Detection (CVPR 2021 · similarity 0.86)
- No Pains, More Gains: Recycling Sub-Salient Patches for Efficient High-Resolution Image Recognition (CVPR 2025 · similarity 0.85)
- SAVSR: Arbitrary-Scale Video Super-Resolution via a Learned Scale-Adaptive Network (AAAI 2024 · similarity 0.85)
- KD-DETR: Knowledge Distillation for Detection Transformer with Consistent Distillation Points Sampling (CVPR 2024 · similarity 0.84)
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.