Research map / Video / Action recognition
Search-Map-Search: A Frame Selection Paradigm for Action 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 |
|---|---|---|---|
| Use SMS-selected frames instead of Base sampling | ResNet-50 action recognition · ActivityNet, 8 frames | mAP: 77.34\pm0.06 -> 83.72\pm0.05 | Helped |
| Use SMS-selected frames instead of Base sampling | ResNet-50 action recognition · FCVID, 8 frames | mAP: 83.84\pm0.05 -> 86.54\pm0.06 | Helped |
| Use SMS-selected frames instead of Base sampling | ResNet-50 action recognition · UCF101, 8 frames | mAP: 90.70\pm0.10 -> 92.25\pm0.10 | Helped |
| Train with SMS-selected frames, in addition to using them at inference | SMS on ActivityNet · ActivityNet, 8 frames | mAP: 82.76\pm0.15 -> 83.72\pm0.05 | Helped |
| Train with SMS-selected frames, in addition to using them at inference | SMS on FCVID · FCVID, 8 frames | mAP: 86.35\pm0.03 -> 86.54\pm0.06 | Helped |
| Use SMS instead of SMART* frame selection | ResNet-50 action recognition · ActivityNet, 8 frames | mAP: 80.67 -> 83.72 | Helped |
| Use SMS instead of SMART* frame selection | ResNet-50 action recognition · FCVID, 8 frames | mAP: 83.35 -> 86.54 | Helped |
| Use a transformer rather than mean-pooling plus a two-layer MLP for feature mapping (ablation) | SMS · ActivityNet, 8 frames | mAP: 83.22 -> 83.72 | Helped |
| Use Kinetics-pretrained rather than ImageNet-pretrained ResNet-50 features (ablation) | SMS · ActivityNet, 8 frames | mAP: 79.97 -> 83.72 | Helped |
| Use Kinetics-pretrained rather than ActivityNet-pretrained ResNet-50 features (ablation) | SMS · ActivityNet, 8 frames | mAP: 81.06 -> 83.72 | Helped |
| Replace ResNet-50 feature extraction with MobileNet-V2 (ablation) | SMS · ActivityNet, 8 frames | mAP: 83.72 -> 79.53 | Hurt |
| Use SMS-selected frames instead of Base sampling | TimeSFormer · ActivityNet, 8 frames | mAP: 90.11 -> 91.97 | Helped |
Closest papers by meaning
- SMART Frame Selection for Action Recognition (AAAI 2021 · similarity 0.89)
- No Frame Left Behind: Full Video Action Recognition (CVPR 2021 · similarity 0.89)
- Attention-Aware Sampling via Deep Reinforcement Learning for Action Recognition (AAAI 2019 · similarity 0.89)
- Action Detail Matters: Refining Video Recognition with Local Action Queries (CVPR 2025 · similarity 0.86)
- FACT: Frame-Action Cross-Attention Temporal Modeling for Efficient Action Segmentation (CVPR 2024 · similarity 0.86)
- Re-thinking Temporal Search for Long-Form Video Understanding (CVPR 2025 · 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.