Research map / Video / Video-language understanding
DeCo: Decomposition and Reconstruction for Compositional Temporal Grounding via Coarse-To-Fine Contrastive Ranking
Lijin Yang, Quan Kong, Hsuan-Kung Yang, Wadim Kehl, Yoichi Sato, Norimasa Kobori
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 coarse-to-fine decomposition, reconstruction, and contrastive ranking (ablation) | DeCo · Charades-CG Novel composition | R1@0.5: 40.41 -> 47.39 | Improved |
| Add sentence recomposition to the full contrastive-ranking configuration (ablation) | DeCo · Charades-CG Novel-Composition | mIoU: 40.14 -> 40.70 | Improved |
| Add ranking constraints for original query, positive sub-sentence, and negative sub-sentences (ablation) | DeCo · Charades-CG Novel-Composition | mIoU: 36.48 -> 40.14 | Improved |
| Add positive sub-sentence without contrastive ranking (ablation) | DeCo · Charades-CG Test-trivial | mIoU: 45.46 -> 45.08 | Hurt |
| Add positive sub-sentence with contrastive ranking (ablation) | DeCo · Charades-CG Test-trivial | mIoU: 45.08 -> 47.05 | Improved |
| Add the cross-sentence negative sub-sentence c to the ranking order (ablation) | DeCo · Charades-CG Novel-Composition | mIoU: 38.29 -> 39.20 | Improved |
| Add the unrelated negative sub-sentence o to the ranking order (ablation) | DeCo · Charades-CG Novel-Composition | mIoU: 38.29 -> 38.65 | Improved |
| Use both negative sub-sentence types in ranking (ablation) | DeCo · Charades-CG Novel-Composition | mIoU: 38.29 -> 39.46 | Improved |
| Use two learnable prompt tokens instead of no learnable prompt tokens (ablation) | DeCo · Charades-CG Novel-Composition | mIoU: 38.57 -> 40.14 | Improved |
| Increase decomposition prompt length from M = 2 to M = 4 (ablation) | DeCo · Charades-CG Novel-Composition | mIoU: 40.14 -> 39.95 | Hurt |
| Increase decomposition prompt length from M = 2 to M = 8 (ablation) | DeCo · Charades-CG Novel-Composition | mIoU: 40.14 -> 39.59 | Hurt |
| Increase recomposition token length from N = 1 to N = 4 while keeping M = 2 (ablation) | DeCo · Charades-CG Novel-Composition | R1@0.5: 45.26 -> 47.39 | Improved |
Later papers that compared with this method
3 later accepted papers report DECO as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- EgoTac: In-the-wild Tactile Prediction from Egocentric Vision (NeurIPS 2026 · better on 4 of 4 shared results)
- DisCo: Towards Distinct and Coherent Visual Encapsulation in Video MLLMs (ICCV 2025 · better on 1 of 1 shared result)
- SHINE: Saliency-aware HIerarchical NEgative Ranking for Compositional Temporal Grounding (ECCV 2024 · 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
- Compositional Temporal Grounding With Structured Variational Cross-Graph Correspondence Learning (CVPR 2022 · similarity 0.92)
- Variational Cross-Graph Reasoning and Adaptive Structured Semantics Learning for Compositional Temporal Grounding (TPAMI 2023 · similarity 0.91)
- Explore Inter-contrast between Videos via Composition for Weakly Supervised Temporal Sentence Grounding (AAAI 2022 · similarity 0.90)
- SnAG: Scalable and Accurate Video Grounding (CVPR 2024 · similarity 0.88)
- Separating the "Chirp" from the "Chat": Self-supervised Visual Grounding of Sound and Language (CVPR 2024 · similarity 0.88)
- Universal Video Temporal Grounding with Generative Multi-modal Large Language Models (NeurIPS 2025 · 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.
