Research map / Video / Action recognition
Weakly-Supervised Temporal Action Detection for Fine-Grained Videos with Hierarchical Atomic Actions
Zhi Li, Lu He, Huijuan Xu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 pseudo-label self-supervised clustering loss (ablation) | HAAN · FineAction val | avg mAP: 2.72 -> 3.18 | Improved |
| add visual concept loss (ablation) | HAAN · FineAction val | avg mAP: 3.18 -> 3.57 | Improved |
| add coarse-level label hierarchy loss (ablation) | HAAN · FineAction val | avg mAP: 3.57 -> 4.10 | Improved |
| add pseudo-label loss (ablation) | HAAN · FineGym-99 (new split) | avg mAP: 5.61 -> 6.09 | Improved |
| add visual concept loss (ablation) | HAAN · FineGym-99 (new split) | avg mAP: 6.09 -> 6.95 | Improved |
| add coarse-level label hierarchy loss (ablation) | HAAN · FineGym-99 (new split) | avg mAP: 6.95 -> 7.67 | Improved |
| use cosine instead of Euclidean distance (ablation) | HAAN · FineAction val | avg mAP: 3.26 -> 3.57 | Improved |
| use K-means instead of GMM clustering (500 clusters) (ablation) | HAAN · FineAction val | mAP@0.5: 6.14 -> 6.40 | Improved |
| use K-means instead of Birch clustering (500 clusters) (ablation) | HAAN · FineAction val | mAP@0.5: 6.31 -> 6.40 | Improved |
| vary number of K-means clusters from 250 to 1000 (ablation) | HAAN · FineAction val | mAP@0.5: 6.31, 6.40, 6.25, 6.26 | No clear effect |
| HAAN vs best weakly-supervised baseline WTAL-UM | HAAN · FineAction val | avg mAP: 3.64 -> 4.10 | Improved |
| HAAN vs best weakly-supervised baseline WTAL-UM | HAAN · FineGym-99 (new split) | avg mAP: 6.11 -> 7.67 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- D2-Net — D2-Net: Weakly-Supervised Action Localization via Discriminative Embeddings and Denoised Activations (ICCV 2021) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- FineGym: A Hierarchical Video Dataset for Fine-Grained Action Understanding (CVPR 2020 · similarity 0.91)
- Action Sets: Weakly Supervised Action Segmentation Without Ordering Constraints (CVPR 2018 · similarity 0.90)
- SeFAR: Semi-supervised Fine-grained Action Recognition with Temporal Perturbation and Learning Stabilization (AAAI 2025 · similarity 0.90)
- Storyboard-guided Alignment for Fine-grained Video Action Recognition (NeurIPS 2025 · similarity 0.90)
- Fine-Grained Temporal Contrastive Learning for Weakly-Supervised Temporal Action Localization (CVPR 2022 · similarity 0.90)
- Action Recognition From Single Timestamp Supervision in Untrimmed Videos (CVPR 2019 · similarity 0.90)
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.
