Research map / Vision-language and multimodal / CLIP and zero-shot
CoTMR: Chain-of-Thought Multi-Scale Reasoning for Training-Free Zero-Shot Composed Image Retrieval
Zelong Sun, Dong Jing, Zhiwu Lu
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 positive object score to base score (ablation) | CoTMR (CLIP ViT-G/14) · Fashion-IQ val | Recall@10: 33.99 -> 35.62 | Improved |
| add negative object score to base score (ablation) | CoTMR (CLIP ViT-G/14) · Fashion-IQ val | Recall@10: 33.99 -> 34.42 | Improved |
| add both positive and negative scores to base (ablation) | CoTMR (CLIP ViT-G/14) · Fashion-IQ val | Recall@10: 33.99 -> 37.72 | Improved |
| add both positive and negative scores to base (ablation) | CoTMR (CLIP ViT-G/14) · CIRCO val | mAP@5: 26.40 -> 28.87 | Improved |
| add base score to object-only scoring (ablation) | CoTMR (CLIP ViT-G/14) · Fashion-IQ val | Recall@10: 30.50 -> 37.72 | Improved |
| use predefined CIRCoT instead of no CoT (ablation) | CoTMR (CLIP ViT-G/14) · Fashion-IQ val | Recall@10: 31.03 -> 33.99 | Improved |
| use predefined CIRCoT instead of no CoT (ablation) | CoTMR (CLIP ViT-G/14) · CIRCO val | mAP@5: 20.07 -> 26.40 | Improved |
| use self-constructed DDCoT reasoning instead of no CoT (ablation) | CoTMR (CLIP ViT-G/14) · Fashion-IQ val | Recall@10: 31.03 -> 29.21 | Hurt |
| add step-wise reasoning examples to CIRCoT (ablation) | CoTMR (CLIP ViT-G/14) · Fashion-IQ val | Recall@10: 33.41 -> 33.99 | Improved |
| add task-1 reference image observation (ablation) | CoTMR (CLIP ViT-G/14) · CIRCO val | mAP@5: 23.15 -> 26.40 | Improved |
| add task-2 modification text analysis (ablation) | CoTMR (CLIP ViT-G/14) · Fashion-IQ val | Recall@10: 32.32 -> 33.99 | Improved |
| add task-3 modification execution (ablation) | CoTMR (CLIP ViT-G/14) · CIRCO val | mAP@5: 21.58 -> 26.40 | Improved |
Later papers that compared with this method
2 later accepted papers report CoTMR as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- WISER: Wider Search, Deeper Thinking, and Adaptive Fusion for Training-Free Zero-Shot Composed Image Retrieval (CVPR 2026 · better on 8 of 8 shared results)
- Generating a Paracosm for Training-Free Zero-Shot Composed Image Retrieval (ECCV 2026 · better on 2 of 2 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
- Reason-before-Retrieve: One-Stage Reflective Chain-of-Thoughts for Training-Free Zero-Shot Composed Image Retrieval (CVPR 2025 · similarity 0.93)
- Vision-by-Language for Training-Free Compositional Image Retrieval (ICLR 2024 · similarity 0.91)
- RMIR: A Benchmark Dataset for Reasoning-Intensive Multimodal Image Retrieval (CVPR 2026 · similarity 0.90)
- CoLLM: A Large Language Model for Composed Image Retrieval (CVPR 2025 · similarity 0.89)
- Context-I2W: Mapping Images to Context-Dependent Words for Accurate Zero-Shot Composed Image Retrieval (AAAI 2024 · similarity 0.89)
- Missing Target-Relevant Information Prediction with World Model for Accurate Zero-Shot Composed Image Retrieval (CVPR 2025 · similarity 0.89)
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.
