Research map / Video / Action recognition
ActionBytes: Learning From Trimmed Videos to Localize Actions
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 ActionByte mining instead of the baseline training without ActionBytes | ActionBytes localization model · Thumos14, ActivityNet1.2, and MultiThumos | mAP at reported IoU thresholds: 8.4 -> 11.3 (Thumos14 IoU=0.5); 19.4 -> 20.1 (ActivityNet1.2 IoU=0.5); 3.2 -> 4.1 (MultiThumos IoU=0.5) | Helped |
| Add ActionByte proposals to the localization detections (ablation) | ActionBytes localization model · Thumos14, ActivityNet1.2, and MultiThumos | mAP at IoU 0.5: 11.3 -> 15.5 (Thumos14); 20.1 -> 20.3 (ActivityNet1.2); 4.1 -> 6.1 (MultiThumos) | Helped |
| Replace fixed-size segments with automatically extracted ActionByte boundaries (ablation) | ActionBytes localization model · Thumos14 | mAP @ IoU 0.5: 14.1% -> 15.5% | Helped |
| Use 500 clusters to generate pseudo-labels (ablation) | ActionBytes mining pipeline · Thumos14 | localization performance: 500 | Mixed |
| Merge up to 4 ActionBytes for proposals instead of using single ActionByte proposals (ablation) | ActionByte proposals · Thumos14 | mAP: +3% | Helped |
| Use ActionByte proposals in weakly supervised localization (ablation) | Paul et al. localization model · Thumos14, ActivityNet1.2, and MultiThumos | mAP at IoU 0.5: 26.5 -> 29.0 (Thumos14); 37.0 -> 39.4 (ActivityNet1.2); 8.9 -> 12.1 (MultiThumos) | Helped |
| Use ActionBytes mining and proposals in zero-shot localization | ActionBytes localization model · Thumos14 and MultiThumos | mAP at IoU 0.5: 3.1 -> 6.3 (Thumos14); 1.3 -> 3.0 (MultiThumos) | Helped |
Closest papers by meaning
- Completeness Modeling and Context Separation for Weakly Supervised Temporal Action Localization (CVPR 2019 · similarity 0.91)
- One-Shot Action Localization by Learning Sequence Matching Network (CVPR 2018 · similarity 0.91)
- ASM-Loc: Action-Aware Segment Modeling for Weakly-Supervised Temporal Action Localization (CVPR 2022 · similarity 0.91)
- PivoTAL: Prior-Driven Supervision for Weakly-Supervised Temporal Action Localization (CVPR 2023 · similarity 0.90)
- Learning Transferable Self-Attentive Representations for Action Recognition in Untrimmed Videos with Weak Supervision (AAAI 2019 · similarity 0.90)
- Weakly-Supervised Temporal Action Localization by Inferring Salient Snippet-Feature (AAAI 2024 · similarity 0.89)
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.