Research map / Video / Video segmentation and tracking
Efficient Video Classification Using Fewer Frames
Check 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 |
|---|---|---|---|
| Distill from a full-frame H-RNN teacher to a student using 30 uniformly spaced frames | H-RNN video classification · YouTube-8M validation | GAP: 0.795 -> 0.806 | Helped |
| Distill from a full-frame H-RNN teacher to a student using 30 uniformly spaced frames | H-RNN video classification · YouTube-8M validation | mAP: 0.378 -> 0.405 | Helped |
| Use equally spaced rather than randomly sampled frames (ablation) | H-RNN frame-limited baseline · YouTube-8M validation, k=6 | GAP: 0.679 -> 0.715 | Helped |
| Use equally spaced rather than first, middle, last-segment, or last frames (ablation) | H-RNN frame-limited baseline · YouTube-8M validation, k=6 | GAP: 0.478 -> 0.715 | Helped |
| Use serial instead of parallel teacher-student training (ablation) | H-RNN student with L_rep + L_pred + L_CE · YouTube-8M validation, k=30 | GAP: 0.806 -> 0.806 | No clear effect |
| Add the prediction-matching loss to representation and classification losses (ablation) | Serial H-RNN student · YouTube-8M validation, k=30 | GAP: 0.803 -> 0.806 | Helped |
| Add representation matching to prediction and classification losses (ablation) | Serial H-RNN student · YouTube-8M validation, k=30 | GAP: 0.799 -> 0.806 | Helped |
| Add intermediate-representation matching to final-representation matching (ablation) | Teacher-Student H-RNN · YouTube-8M validation, k=30 | GAP: 0.806 -> 0.806 | No clear effect |
| Distill to a NetVLAD student using 10 frames rather than training the Uniform baseline | NetVLAD video classification · YouTube-8M validation, k=10 | mAP: 0.364 -> 0.383 | Helped |
| Distill to a NeXtVLAD student using 30 frames rather than training the Uniform baseline | NeXtVLAD video classification · YouTube-8M validation, k=30 | mAP: 0.424 -> 0.439 | Helped |
| Distill to a NeXtVLAD student using 30 frames rather than training the Uniform baseline | NeXtVLAD video classification · YouTube-8M validation, k=30 | GAP: 0.812 -> 0.818 | Helped |
| Reduce the input frames from all frames to 30 for NeXtVLAD | NeXtVLAD student · YouTube-8M validation | FLOPs (in Billion): 1.337 -> 0.134 | Helped |
Closest papers by meaning
- FASTER Recurrent Networks for Efficient Video Classification (AAAI 2020 · similarity 0.88)
- Tensor-Train Recurrent Neural Networks for Video Classification (ICML 2017 · similarity 0.87)
- More Is Less: Learning Efficient Video Representations by Big-Little Network and Depthwise Temporal Aggregation (NeurIPS 2019 · similarity 0.86)
- Recurrent Residual Module for Fast Inference in Videos (CVPR 2018 · similarity 0.85)
- Learning a Condensed Frame for Memory-Efficient Video Class-Incremental Learning (NeurIPS 2022 · similarity 0.85)
- Dancing with Still Images: Video Distillation via Static-Dynamic Disentanglement (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.