Research map / Video / Action recognition
Weakly-Supervised Action Localization with Expectation-Maximization Multi-Instance Learning
Zhekun Luo, Devin Guillory, Baifeng Shi, Wei Ke, Fang Wan, Trevor Darrell, 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 the proposed pseudo labeling scheme to alternating training (ablation) | EM-MIL · THUMOS14 | mAP@0.5 (%): 24.5 -> 30.5 | Improved |
| add EM alternating training to pseudo labeling (instead of joint optimization) (ablation) | EM-MIL · THUMOS14 | mAP@0.5 (%): 26.8 -> 30.5 | Improved |
| use UntrimmedNet features instead of I3D two-stream features | EM-MIL · THUMOS14 | mAP@0.5 (%): 30.5 -> 27.2 | Hurt |
| vary threshold gamma in [0.05, 0.2] (ablation) | EM-MIL · THUMOS14 | mAP@0.5 (%): 29.0 to 30.5 | No clear effect |
| use EM-MIL instead of previous best I3D-based weakly supervised model | EM-MIL · THUMOS14 | mAP@0.5 (%): 26.8 -> 30.5 | Improved |
| use EM-MIL instead of Nguyen et al. at low tIoU | EM-MIL · THUMOS14 | mAP@0.1 (%): 64.2 -> 59.1 | Hurt |
| use EM-MIL instead of Nguyen et al. at high tIoU | EM-MIL · THUMOS14 | mAP@0.7 (%): 8.6 -> 16.4 | Improved |
| use EM-MIL instead of 3C-Net at high tIoU | EM-MIL · ActivityNet1.2 | mAP@0.9 (%): 9.2 -> 2.0 | Hurt |
| use EM-MIL instead of 3C-Net at tIoU 0.5 | EM-MIL · ActivityNet1.2 | mAP@0.5 (%): 37.2 -> 37.4 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- 3C-Net — 3C-Net: Category Count and Center Loss for Weakly-Supervised Action Localization (ICCV 2019) · better on 2 of 5 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
- Weakly-Supervised Action Localization by Hierarchically-Structured Latent Attention Modeling (ICCV 2023 · similarity 0.93)
- Proposal-Based Multiple Instance Learning for Weakly-Supervised Temporal Action Localization (CVPR 2023 · similarity 0.93)
- ASM-Loc: Action-Aware Segment Modeling for Weakly-Supervised Temporal Action Localization (CVPR 2022 · similarity 0.91)
- Weakly-Supervised Action Localization With Background Modeling (ICCV 2019 · similarity 0.90)
- Bridge the Gap: From Weak to Full Supervision for Temporal Action Localization with PseudoFormer (CVPR 2025 · similarity 0.89)
- Weakly-Supervised Temporal Action Localization by Inferring Salient Snippet-Feature (AAAI 2024 · similarity 0.88)
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.
