Research map / Video / Action recognition
3C-Net: Category Count and Center Loss for Weakly-Supervised Action Localization
Sanath Narayan, Hisham Cholakkal, Fahad Shahbaz Khan, Ling Shao
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 multi-label center loss (ablation) | CLS baseline · THUMOS14 localization | mAP@IoU0.5: 19.1 -> 24.6 | Improved |
| add counting loss (ablation) | CLS + CL · THUMOS14 localization | mAP@IoU0.5: 24.6 -> 26.6 | Improved |
| add center loss (ablation) | CLS baseline · ActivityNet 1.2 localization | mean mAP: 18.2 -> 21.1 | Improved |
| add counting loss (ablation) | CLS + CL · ActivityNet 1.2 localization | mean mAP: 21.1 -> 21.7 | Improved |
| add classification loss in both streams (ablation) | 3C-Net · THUMOS14 localization | mAP@IoU0.5: 2.5% mAP drop when removed | Improved |
| add center loss in both RGB and flow streams (vs flow only) (ablation) | 3C-Net · THUMOS14 localization | mAP@IoU0.5: 2.1% mAP drop when flow only | Improved |
| add center loss in both streams (vs RGB only) (ablation) | 3C-Net · THUMOS14 localization | mAP@IoU0.5: 1.9% mAP drop when RGB only | Improved |
| add negative category counting loss (ablation) | 3C-Net · THUMOS14 localization | mAP@IoU0.5: 1.5% mAP drop when removed | Improved |
| use relative error rather than absolute error in counting loss (ablation) | 3C-Net · THUMOS14 localization | mAP@IoU0.5: 1.2% mAP drop with absolute error | Improved |
| 3C-Net vs W-TALC | W-TALC · THUMOS14 localization | mAP@IoU0.5: 22.0 -> 26.6 | Improved |
| 3C-Net vs W-TALC | W-TALC · ActivityNet 1.2 localization | mean mAP: 18.0 -> 21.7 | Improved |
| 3C-Net vs W-TALC | W-TALC · ActivityNet 1.2 classification | mAP: 93.2 vs 92.4 | No clear effect |
Later papers that compared with this method
1 later accepted paper reports 3C-Net as a baseline in their results tables; 0 of them report a better number on most of the results they share with it.
- Weakly-Supervised Action Localization with Expectation-Maximization Multi-Instance Learning (ECCV 2020 · better on 2 of 5 shared results)
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.
- R-C3D — R-C3D: Region Convolutional 3D Network for Temporal Activity Detection (ICCV 2017) · better on 0 of 1 shared result
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
- D2-Net: Weakly-Supervised Action Localization via Discriminative Embeddings and Denoised Activations (ICCV 2021 · similarity 0.91)
- Action Unit Memory Network for Weakly Supervised Temporal Action Localization (CVPR 2021 · similarity 0.91)
- Relational Prototypical Network for Weakly Supervised Temporal Action Localization (AAAI 2020 · similarity 0.90)
- Two-Branch Relational Prototypical Network for Weakly Supervised Temporal Action Localization (TPAMI 2021 · similarity 0.90)
- Weakly-Supervised Temporal Action Localization by Inferring Salient Snippet-Feature (AAAI 2024 · similarity 0.90)
- Similar Modality Enhancement and Action Consistency Learning for Weakly Supervised Temporal Action Localization (AAAI 2025 · 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.
