Research map / Video / Action recognition
UntrimmedNets for Weakly Supervised Action Recognition and Detection
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 shot-based rather than uniform clip sampling (ablation) | UntrimmedNet with soft selection · THUMOS14 | mAP: 70.2% vs. 71.6% | Helped |
| Use Temporal Segment Networks rather than the original Two-Stream CNN for feature extraction (ablation) | UntrimmedNet with soft selection · THUMOS14 | mAP: 71.6% vs. 74.2% | Helped |
| Increase TSN training segments from 3 to 21 (ablation) | TSN · THUMOS14 test (WSR) | mAP: 67.7% -> 68.5% | Helped |
| Add a hard selection module | TSN-based model · THUMOS14 test (WSR) | mAP: 67.7% -> 73.6% | Helped |
| Add a soft selection module | TSN-based model · THUMOS14 test (WSR) | mAP: 67.7% -> 74.2% | Helped |
| Increase sampled proposals per video from 5 to 7 or 9 (ablation) | UntrimmedNet with soft selection · THUMOS14 | mAP | Helped |
| Increase sampled proposals per video (ablation) | UntrimmedNet with hard selection · THUMOS14 | mAP | No clear effect |
| Use hard rather than soft selection (ablation) | UntrimmedNet · ActivityNet v1.2 validation (WSR, setting a) | mAP: 87.7% vs. 86.9% | Mixed |
Closest papers by meaning
- Learning Transferable Self-Attentive Representations for Action Recognition in Untrimmed Videos with Weak Supervision (AAAI 2019 · similarity 0.91)
- Action Recognition From Single Timestamp Supervision in Untrimmed Videos (CVPR 2019 · similarity 0.89)
- WOAD: Weakly Supervised Online Action Detection in Untrimmed Videos (CVPR 2021 · similarity 0.88)
- Weakly Supervised Action Learning With RNN Based Fine-To-Coarse Modeling (CVPR 2017 · similarity 0.88)
- Action Unit Memory Network for Weakly Supervised Temporal Action Localization (CVPR 2021 · similarity 0.88)
- Unsupervised Learning of Action Classes With Continuous Temporal Embedding (CVPR 2019 · 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.