Research map / Video / Action recognition
D2-Net: Weakly-Supervised Action Localization via Discriminative Embeddings and Denoised Activations
Sanath Narayan, Hisham Cholakkal, Munawar Hayat, Fahad Shahbaz Khan, Ming-Hsuan Yang, 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 focal loss instead of cross-entropy (ablation) | D2-Net baseline · THUMOS14 test | mAP@IoU0.5: 23.0 -> 26.7 | Improved |
| add discriminative loss L_Dis instead of focal loss (ablation) | D2-Net · THUMOS14 test | mAP@IoU0.5: 26.7 -> 32.2 | Improved |
| add denoising loss L_D to L_Dis (ablation) | D2-Net · THUMOS14 test | mAP@IoU0.5: 32.2 -> 36.0 | Improved |
| add denoising loss L_D to L_Dis (ablation) | D2-Net · THUMOS14 test | F1@IoU0.5: 30.7 -> 36.7 | Improved |
| add grouping and clustering weights to the focal penalty (ablation) | L_Dis · THUMOS14 test | mAP@IoU0.5: 29.2 -> 32.2 | Improved |
| use MI-based pDMI loss instead of L1 for denoising (ablation) | D2-Net · THUMOS14 test | mAP@IoU0.5: 32.9 -> 36.0 | Improved |
| use MI-based pDMI loss instead of BCE for denoising (ablation) | D2-Net · THUMOS14 test | mAP@IoU0.5: 33.5 -> 36.0 | Improved |
| add snippet-level denoising to L_Dis (ablation) | D2-Net · THUMOS14 test | mAP@IoU0.5: 32.2 -> 34.3 | Improved |
| add video-level denoising to L_Dis (ablation) | D2-Net · THUMOS14 test | mAP@IoU0.5: 32.2 -> 33.2 | Improved |
| add both snippet-level and video-level denoising (ablation) | D2-Net · THUMOS14 test | mAP@IoU0.5: 34.3 -> 36.0 | Improved |
| add intra-class grouping weight gamma=0.01 instead of 0 (ablation) | D2-Net · THUMOS14 test | mAP@IoU0.5: 34.2 -> 36.0 | Improved |
| raise intra-class grouping weight gamma from 0.01 to 1.0 (ablation) | D2-Net · THUMOS14 test | mAP@IoU0.5: 36.0 -> 33.7 | Hurt |
Later papers that compared with this method
2 later accepted papers report D2-Net as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Weakly-Supervised Temporal Action Detection for Fine-Grained Videos with Hierarchical Atomic Actions (ECCV 2022 · better on 3 of 3 shared results)
- Exploring Denoised Cross-Video Contrast for Weakly-Supervised Temporal Action Localization (CVPR 2022 · better on 0 of 1 shared result)
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.
- CoLA — CoLA: Weakly-Supervised Temporal Action Localization With Snippet Contrastive Learning (CVPR 2021) · better on 2 of 2 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
- Temporal Action Localization in the Deep Learning Era: A Survey (TPAMI 2023 · similarity 0.92)
- Weakly-Supervised Action Localization With Background Modeling (ICCV 2019 · similarity 0.92)
- PivoTAL: Prior-Driven Supervision for Weakly-Supervised Temporal Action Localization (CVPR 2023 · similarity 0.92)
- Fine-Grained Temporal Contrastive Learning for Weakly-Supervised Temporal Action Localization (CVPR 2022 · similarity 0.92)
- Bridge the Gap: From Weak to Full Supervision for Temporal Action Localization with PseudoFormer (CVPR 2025 · similarity 0.91)
- Weakly-Supervised Temporal Action Localization by Inferring Salient Snippet-Feature (AAAI 2024 · similarity 0.91)
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.
