Research map / Video / Video-language understanding
MomentDiff: Generative Video Moment Retrieval from Random to Real
Pandeng Li, Chen-Wei Xie, Hongtao Xie, Liming Zhao, Lei Zhang, Yun Zheng, Deli Zhao, Yongdong Zhang
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 random-span diffusion denoising instead of learned proposals | MomentDiff · Charades-STA-Len and Charades-STA-Mom | R1@0.5: 34.39 -> 38.32; 22.16 -> 33.59 | Improved |
| Replace FC span embedding with ROI pooling (ablation) | MomentDiff · Charades-STA | R1@0.5: 55.57 -> 50.21 | Hurt |
| Set diffusion scale to 1.0 instead of 2.0 (ablation) | MomentDiff · Charades-STA | R1@0.5: 55.57 -> 54.33 | Hurt |
| Set diffusion scale to 3.0 instead of 2.0 (ablation) | MomentDiff · Charades-STA | R1@0.5: 55.57 -> 50.12 | Hurt |
| Add the VMD denoiser and diffusion process (ablation) | MomentDiff · Charades-STA | R1@0.5: 46.69 -> 55.57 | Improved |
| Add explicit noise-intensity conditioning (ablation) | MomentDiff · Charades-STA | R1@0.5: 50.41 -> 55.57 | Improved |
| Add the VMR localization loss alongside similarity loss (ablation) | MomentDiff · Charades-STA | R1@0.5: 31.53 -> 55.57 | Improved |
| Add similarity loss alongside VMR loss (ablation) | MomentDiff · Charades-STA | R1@0.5: 40.94 -> 55.57 | Improved |
| Use 5 random spans rather than 1, 10, or 20 (ablation) | MomentDiff · Charades-STA | R1@0.5: 50.83; 55.57; 53.89; 50.42 | Mixed |
| Increase diffusion sampling steps (ablation) | MomentDiff · Charades-STA | R1@0.7: 29.78 -> 32.42 | Improved |
| Use DDIM instead of DDPM sampling (ablation) | MomentDiff · Charades-STA | MAP_avg: 32.48 -> 32.85 | Improved |
| Use a cosine rather than linear beta schedule (ablation) | MomentDiff · Charades-STA | R1@0.5: 54.76 -> 55.57 | Improved |
Later papers that compared with this method
1 later accepted paper reports MomentDiff as a baseline in their results tables; 0 of them report a better number on most of the results they share with it.
- Adaptive Evidential Learning for Temporal-Semantic Robustness in Moment Retrieval (AAAI 2026 · better on 1 of 2 shared results)
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.
- RaNet — RANet: Ranking Attention Network for Fast Video Object Segmentation (ICCV 2019) · 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
- Generative Video Diffusion for Unseen Novel Semantic Video Moment Retrieval (AAAI 2025 · similarity 0.91)
- Transferable Video Moment Localization by Moment-Guided Query Prompting (AAAI 2024 · similarity 0.90)
- MomentSeeker: A Task-Oriented Benchmark For Long-Video Moment Retrieval (NeurIPS 2025 · similarity 0.89)
- Fewer Steps, Better Performance: Efficient Cross-Modal Clip Trimming for Video Moment Retrieval Using Language (AAAI 2024 · similarity 0.88)
- When One Moment Isn't Enough: Multi-Moment Retrieval with Cross-Moment Interactions (NeurIPS 2025 · similarity 0.88)
- Anchor-Aware Similarity Cohesion in Target Frames Enables Predicting Temporal Moment Boundaries in 2D (CVPR 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.
