Research map / Video / Action recognition
UCF-STAR: A Large Scale Still Image Dataset for Understanding Human Actions
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 |
|---|---|---|---|
| add the predicted dynamic-skeleton temporal stream to the spatial stream (ablation) | TSSTN · UCF-STAR test | mAP (%): 26.3 -> 91.9 | Helped |
| add the spatial stream to the temporal stream (ablation) | TSSTN · UCF-STAR test | mAP (%): 86.2 -> 91.9 | Helped |
| fine-tune only top 7 layers instead of all layers in the temporal stream (ablation) | temporal stream · UCF-STAR test | mAP (%): 68.3 -> 86.2 | Helped |
| fine-tune a pretrained network instead of training the temporal stream from scratch (ablation) | temporal stream · UCF-STAR test | mAP (%): 61.3 -> 86.2 | Helped |
| use custom top-k loss instead of default softmax loss for ds prediction (ablation) | dynamic-skeleton predictor · predicted ds mAP, 500 epochs | mAP: 76% -> 83% | Helped |
| fine-tune VGG instead of training ds predictor from scratch (ablation) | dynamic-skeleton predictor · predicted ds mAP, 500 epochs | mAP: 65% -> 76% | Helped |
| TSSTN vs prior still-image methods | TSSTN · UCF-STAR test | mAP (%): 70.9 -> 91.9 | Helped |
| TSSTN vs best prior on Stanford-40 (all classes) | TSSTN · Stanford-40 (All) | mAP (%): 90.9 -> 86.3 | Hurt |
| TSSTN on body-motion vs non-body-motion Stanford-40 classes compared with Gkioxari et al. | TSSTN · Stanford-40 | mAP (%): 93.87 -> 97.8 (body-motion); 89.73 -> 80.2 (non-body-motion) | Mixed |
| TSSTN vs best prior on Willow body-motion | TSSTN · Willow overall | mAP (%): 79.3 -> 84.6 | Helped |
| TSSTN vs best prior on WIDER | TSSTN · WIDER body-motion | mAP (%): 86.2 -> 90.3 | Helped |
Closest papers by meaning
- Taylor Videos for Action Recognition (ICML 2024 · similarity 0.87)
- VA-AR: Learning Velocity-Aware Action Representations with Mixture of Window Attention (AAAI 2025 · similarity 0.86)
- Part-aware Unified Representation of Language and Skeleton for Zero-shot Action Recognition (CVPR 2024 · similarity 0.86)
- H-MoRe: Learning Human-centric Motion Representation for Action Analysis (CVPR 2025 · similarity 0.86)
- Temporal Alignment-Free Video Matching for Few-shot Action Recognition (CVPR 2025 · similarity 0.86)
- ActionAtlas: A VideoQA Benchmark for Domain-specialized Action Recognition (NeurIPS 2024 · similarity 0.86)
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.