Research map / Video / Action recognition
DDG-Net: Discriminability-Driven Graph Network for Weakly-supervised Temporal Action Localization
Xiaojun Tang, Junsong Fan, Chuanchen Luo, Zhaoxiang Zhang, Man Zhang, Zongyuan Yang
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 DDG-Net (graph with discrimination-driven connections and feature consistency loss) to base WTAL model (ablation) | base WTAL model · THUMOS14 | mAP Avg (%): 45.8 -> 47.3 | Improved |
| add graph average only to the baseline (ablation) | base WTAL model · THUMOS14 | mAP Avg (%): 45.8 -> 46.4 | Improved |
| add graph convolution without feature consistency loss (ablation) | base WTAL model with graph average · THUMOS14 | mAP Avg (%): 46.4 -> 36.1 | Hurt |
| add feature consistency loss to graph convolution (ablation) | DDG-Net · THUMOS14 | mAP Avg (%): 36.1 -> 47.3 | Improved |
| use fused adjacency matrix across RGB and flow instead of separate adjacency matrices (ablation) | DDG-Net · THUMOS14 | mAP Avg (%): 44.9 -> 47.3 | Improved |
| disconnect ambiguous snippets so they only receive information (ablation) | DDG-Net · THUMOS14 | mAP Avg (%): 45.8 -> 47.3 | Improved |
| raise similarity threshold from 0.8 to 0.9 (ablation) | DDG-Net · THUMOS14 | mAP | Hurt |
| choose action threshold 0.5 versus higher values (ablation) | DDG-Net · THUMOS14 | mAP | Improved |
| add DDG-Net on ActivityNet1.2 | base WTAL model · ActivityNet1.2 | mAP | Improved |
| add DDG-Net to base model | base WTAL model · THUMOS14 | parameters and FLOPs: 21.5 M -> 29.9 M params; 6.89 G -> 13.39 G FLOPs | Hurt |
Later papers that compared with this method
2 later accepted papers report DDG-Net as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- Proposal Score Realignment Guided by Semantic Completeness for Weakly Supervised Temporal Action Localization (ECCV 2026 · better on 3 of 3 shared results)
- Face-Guided Sentiment Boundary Enhancement for Weakly-Supervised Temporal Sentiment Localization (CVPR 2026 · better on 1 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 0.90)
- ACGNet: Action Complement Graph Network for Weakly-Supervised Temporal Action Localization (AAAI 2022 · similarity 0.90)
- Distilling Vision-Language Pre-Training To Collaborate With Weakly-Supervised Temporal Action Localization (CVPR 2023 · similarity 0.89)
- 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)
- Similar Modality Enhancement and Action Consistency Learning for Weakly Supervised Temporal Action Localization (AAAI 2025 · 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.
