Research map / Video / Action recognition
No Frame Left Behind: Full Video Action 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 |
|---|---|---|---|
| Use full-video training with cumulative temporal clustering | TSM-ResNet50 · Breakfast | Top-1 accuracy: 72.1% -> 76.6% | Helped |
| Use full-video training with cumulative temporal clustering | TSM-ResNet50 · Something-Something V1 and V2 | Top-1 accuracy: 0.3% and 0.4% | Helped |
| Use full-video training with cumulative temporal clustering | TSM-ResNet50 · UCF-101 | Top-1 accuracy | Helped |
| Use full-video training with cumulative temporal clustering | TSM-ResNet50 · HMDB51 | Top-1 accuracy | Hurt |
| Replace adaptive clustering with uniform temporal grouping (ablation) | Full-video action recognition · Move4MNIST test | Top-1 accuracy: 94.08 ± 0.25 -> 90.25 ± 0.28 | Hurt |
| Replace adaptive clustering with uniform temporal grouping (ablation) | Full-video action recognition · Breakfast | Top-1 accuracy: 62.0 -> 58.3 | Hurt |
| Use cumulative clustering instead of slope clustering (ablation) | Full-video action recognition · Move4MNIST test | Top-1 accuracy: 93.33 ± 0.19 -> 94.08 ± 0.25 | Helped |
| Use cumulative clustering instead of slope clustering (ablation) | Full-video action recognition · Breakfast | Top-1 accuracy: 60.7 -> 62.0 | Helped |
| Increase the cluster count from 8 to 16 (ablation) | Full-video action recognition · Breakfast | Top-1 accuracy: 62.0 -> 64.4 | Helped |
| Increase the cluster count from 8 to 16 (ablation) | Full-video action recognition · Something-Something V1 and V2 | Top-1 accuracy: 49.5% -> 51.4%; 62.7% -> 65.1% | Helped |
| Use cumulative clustering instead of slope clustering (ablation) | Full-video action recognition · Something-Something V1 and V2 | Top-1 accuracy: 46.7% -> 49.5%; 60.2% -> 62.7% | Helped |
| Train on all frames instead of 8 or 16 sampled frames | TSM · Move4MNIST test | Top-1 accuracy: 90.13 ± 0.38; 93.78 ± 0.33 -> 98.83 ± 0.16 | Helped |
Closest papers by meaning
- More Is Less: Learning Efficient Video Representations by Big-Little Network and Depthwise Temporal Aggregation (NeurIPS 2019 · similarity 0.90)
- Large-Scale Weakly-Supervised Pre-Training for Video Action Recognition (CVPR 2019 · similarity 0.90)
- Search-Map-Search: A Frame Selection Paradigm for Action Recognition (CVPR 2023 · similarity 0.89)
- Unsupervised Deep Learning of Mid-Level Video Representation for Action Recognition (AAAI 2018 · similarity 0.89)
- Action Detail Matters: Refining Video Recognition with Local Action Queries (CVPR 2025 · similarity 0.88)
- FACT: Frame-Action Cross-Attention Temporal Modeling for Efficient Action Segmentation (CVPR 2024 · similarity 0.87)
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.