Research map / Video / Action recognition
DiffTAD: Temporal Action Detection with Proposal Denoising Diffusion
Sauradip Nag, Xiatian Zhu, Jiankang Deng, Yi-Zhe Song, Tao Xiang
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 |
|---|---|---|---|
| use cross-timestep selective conditioning instead of feature concatenation (ablation) | DiffTAD · THUMOS14 | average mAP: 65.9 -> 68.0 | Improved |
| use selective conditioning instead of DiffusionDet proposal renewal (ablation) | DiffTAD · THUMOS14 | average mAP: 65.1 -> 68.0 | Improved |
| combine DDIM denoising with selective conditioning (ablation) | DiffTAD · THUMOS14 | average mAP (step 10): 64.9 -> 68.0 | Improved |
| increase sampling steps with no denoising and no selective conditioning (ablation) | DiffTAD · THUMOS14 | average mAP: 62.7 -> 61.3 (step 1 to step 10) | Hurt |
| use late fusion with decoupled RGB and flow denoising instead of early fusion (ablation) | DiffTAD · THUMOS14 | average mAP: 66.0 -> 68.0 | Improved |
| set signal scaling to 0.5 rather than the object detection value of 2.0 (ablation) | DiffTAD · THUMOS14 | average mAP: 64.2 -> 68.0 | Improved |
| increase number of proposals from 50 to 100 (ablation) | DiffTAD · THUMOS14 | average mAP: 68.5 -> 68.4 | No clear effect |
| increase sampling steps from 1 to 5 with 50 proposals (ablation) | DiffTAD · THUMOS14 | average mAP: 64.9 -> 68.5 | Improved |
| corrupt labels instead of proposals in the forward process (ablation) | DiffTAD · THUMOS14 | average mAP: 68.0 -> 62.4 | Hurt |
| also corrupt labels in addition to proposals (ablation) | DiffTAD · THUMOS14 | average mAP: 68.0 -> 67.3 | Hurt |
| add NMS post-processing (ablation) | DiffTAD · THUMOS14 | AR@50: 63.6 -> 64.3 | No clear effect |
| add NMS post-processing (ablation) | BMN · THUMOS14 | AR@50: 29.04 -> 32.73 | Improved |
Later papers that compared with this method
1 later accepted paper reports DiffTAD as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- DiffDVC: Accurate Event Detection for Dense Video Captioning via Diffusion Models (AAAI 2025 · better on 1 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.
- ActionFormer — ActionFormer: Localizing Moments of Actions with Transformers (ECCV 2022) · better on 3 of 4 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
- Action Detection via an Image Diffusion Process (CVPR 2024 · similarity 0.89)
- Dual DETRs for Multi-Label Temporal Action Detection (CVPR 2024 · similarity 0.87)
- Emergent Temporal Correspondences from Video Diffusion Transformers (NeurIPS 2025 · similarity 0.87)
- TE-TAD: Towards Full End-to-End Temporal Action Detection via Time-Aligned Coordinate Expression (CVPR 2024 · similarity 0.87)
- DiffDVC: Accurate Event Detection for Dense Video Captioning via Diffusion Models (AAAI 2025 · similarity 0.87)
- Masked Diffusion Vision-Language Models for Temporal Action Localization (NeurIPS 2026 · similarity 0.87)
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.
