Research map / Label-efficient and robust learning / Recognition and long-tail
Augmenting Moment Retrieval: Zero-Dependency Two-Stage Learning
Zhengxuan Wei, Jiajin Tang, Sibei 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 |
|---|---|---|---|
| train on Splice augmented data in a single stage (ablation) | baseline M-DETR variant · QVHighlights val | mAP avg: 44.82 -> 42.56 | Hurt |
| add Boost to Splice in single-stage training (ablation) | Splice only · QVHighlights val | mAP avg: 42.56 -> 43.88 | Improved |
| two-stage training on real data only, no augmentation (ablation) | baseline · QVHighlights val | mAP avg: 44.82 -> 44.66 | No clear effect |
| add Splice augmentation to vanilla two-stage training (ablation) | two-stage · QVHighlights val | mAP avg: 44.66 -> 47.11 | Improved |
| add Boost on top of Splice with vanilla two-stage (ablation) | two-stage · QVHighlights val | mAP avg: 47.11 -> 48.07 | Improved |
| add dual-path distillation to Splice two-stage (ablation) | Splice+two-stage · QVHighlights val | mAP avg: 47.11 -> 49.11 | Improved |
| add discriminative contrastive loss to Splice two-stage (ablation) | Splice+two-stage · QVHighlights val | mAP avg: 47.11 -> 47.87 | Improved |
| add DCL to Splice+Dill (ablation) | Splice+Dill · QVHighlights val | mAP avg: 49.11 -> 50.64 | Improved |
| add Boost to Splice+Dill+DCL (full model) (ablation) | AMR · QVHighlights val | mAP avg: 50.64 -> 51.66 | Improved |
| add Boost to Splice+Dill (ablation) | Splice+Dill · QVHighlights val | mAP avg: 49.11 -> 50.03 | Improved |
| full AMR versus BAM-DETR | AMR · QVHighlights val | R1@0.5: 65.10 -> 70.13 | Improved |
| full AMR versus UVCOM | AMR · TACoS | R1@0.7: 23.32 -> 26.07 | Improved |
Closest papers by meaning
- When One Moment Isn't Enough: Multi-Moment Retrieval with Cross-Moment Interactions (NeurIPS 2025 · similarity 0.86)
- Transferable Video Moment Localization by Moment-Guided Query Prompting (AAAI 2024 · similarity 0.86)
- Fewer Steps, Better Performance: Efficient Cross-Modal Clip Trimming for Video Moment Retrieval Using Language (AAAI 2024 · similarity 0.85)
- Anchor-Aware Similarity Cohesion in Target Frames Enables Predicting Temporal Moment Boundaries in 2D (CVPR 2025 · similarity 0.85)
- Diversifying Query: Region-Guided Transformer for Temporal Sentence Grounding (AAAI 2025 · similarity 0.85)
- Generative Video Diffusion for Unseen Novel Semantic Video Moment Retrieval (AAAI 2025 · similarity 0.85)
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.
