Research map / Video / Action recognition
VideoLT: Large-Scale Long-Tailed Video Recognition
Xing Zhang, Zuxuan Wu, Zejia Weng, Huazhu Fu, Jingjing Chen, Yu-Gang Jiang, Larry S. Davis
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 |
|---|---|---|---|
| use running AP instead of constant beta=0.5 to set frames per class (ablation) | FrameStack (ResNet-101, nonlinear) · VideoLT | mAP (overall): 0.414 -> 0.532 | Improved |
| use running AP instead of class frequency to set frames per class (ablation) | FrameStack (ResNet-101, nonlinear) · VideoLT | mAP (overall): 0.520 -> 0.532 | Improved |
| use running AP instead of class frequency to set frames per class (ablation) | FrameStack (ResNet-101, NetVLAD) · VideoLT | mAP (overall): 0.663 -> 0.670 | Improved |
| use running AP instead of constant beta=0.5 (ablation) | FrameStack (ResNet-101, NetVLAD) · VideoLT | mAP (overall): 0.648 -> 0.670 | Improved |
| add FrameStack to the nonlinear model on ResNet-101 features | nonlinear classifier · VideoLT | mAP (overall): 0.516 -> 0.532 | Improved |
| add FrameStack to the nonlinear model on ResNet-101 features | nonlinear classifier · VideoLT tail classes | mAP (tail): 0.396 -> 0.417 | Improved |
| add FrameStack with NetVLAD on TSM features | NetVLAD classifier · VideoLT | mAP (overall): 0.660 -> 0.667 | Improved |
| add mixup in feature space | nonlinear classifier (ResNet-101) · VideoLT | mAP (overall): 0.516 -> 0.495 | Hurt |
| add class-balanced sampling | nonlinear classifier (ResNet-101) · VideoLT | mAP (overall): 0.516 -> 0.507 | Hurt |
| add LDAM + DRW | nonlinear classifier (ResNet-101) · VideoLT | mAP (overall): 0.516 -> 0.518 | No clear effect |
| use TSM Kinetics-pretrained features instead of ImageNet ResNet-101 features | FrameStack nonlinear · VideoLT | mAP (overall): 53.2 -> 58.0 | Improved |
| use focal loss instead of BCE in FrameStack (ablation) | FrameStack (ResNet-101, nonlinear) · VideoLT | mAP (overall): 0.521 -> 0.532 | Improved |
Closest papers by meaning
- Minority-Oriented Vicinity Expansion with Attentive Aggregation for Video Long-Tailed Recognition (AAAI 2023 · similarity 0.91)
- MEID: Mixture-of-Experts with Internal Distillation for Long-Tailed Video Recognition (AAAI 2023 · similarity 0.91)
- Use Your Head: Improving Long-Tail Video Recognition (CVPR 2023 · similarity 0.91)
- SimLTD: Simple Supervised and Semi-Supervised Long-Tailed Object Detection (CVPR 2025 · similarity 0.90)
- Deep Long-Tailed Learning: A Survey (TPAMI 2023 · similarity 0.90)
- Large-Scale Long-Tailed Recognition in an Open World (CVPR 2019 · similarity 0.90)
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.
