Research map / Video / Action recognition
AdaFrame: Adaptive Frame Selection for Fast Video Recognition
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 |
|---|---|---|---|
| Adaptively select frames and stop using predicted future utility | ResNet-101 video recognition · FCVID | mAP: 80.2 vs 80.0 | Helped |
| Adaptively select frames and stop using predicted future utility | ResNet-101 video recognition · ActivityNet validation | mAP: 71.5 vs 70.0 | Helped |
| Add global memory with 16 downsampled frames (ablation) | AdaFrame · FCVID | mAP: 77.9 -> 80.2 | Helped |
| Use 12 rather than 16 global-memory frames (ablation) | AdaFrame · FCVID | mAP: 80.2 -> 79.2 | Hurt |
| Use 32 rather than 16 global-memory frames (ablation) | AdaFrame · FCVID | mAP: 80.2 -> 80.2 | No clear effect |
| Replace the margin-based transition reward with ground-truth-class confidence reward (ablation) | AdaFrame · FCVID | mAP: 80.2 -> 78.7 | Hurt |
| Replace the margin-based transition reward with ground-truth confidence transition reward (ablation) | AdaFrame · FCVID | mAP: 80.2 -> 78.9 | Hurt |
| Use predicted utility rather than prediction entropy to decide when to stop (ablation) | AdaFrame adaptive inference · FCVID | adaptive inference across thresholds | Helped |
| Reduce the inference threshold μ (ablation) | AdaFrame · FCVID and ActivityNet | mAP | Mixed |
Closest papers by meaning
- SMART Frame Selection for Action Recognition (AAAI 2021 · similarity 0.90)
- Search-Map-Search: A Frame Selection Paradigm for Action Recognition (CVPR 2023 · similarity 0.88)
- FrameExit: Conditional Early Exiting for Efficient Video Recognition (CVPR 2021 · similarity 0.88)
- AdaFocus V2: End-to-End Training of Spatial Dynamic Networks for Video Recognition (CVPR 2022 · similarity 0.87)
- LookWhen? Fast Video Recognition by Learning When, Where, and What to Compute (NeurIPS 2026 · similarity 0.86)
- LookWhere? Efficient Visual Recognition by Learning Where to Look and What to See from Self-Supervision (NeurIPS 2025 · 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.