Research map / Vision-language and multimodal / CLIP and zero-shot
WISER: Wider Search, Deeper Thinking, and Adaptive Fusion for Training-Free Zero-Shot Composed Image Retrieval
Tianyue Wang, Leigang Qu, Tianyu Yang, Xiangzhao Hao, Yifan Xu, Haiyun Guo, Jinqiao Wang
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 adaptive verifier-based fusion instead of fixed-weight averaging (ablation) | WISER T2I+I2I · CIRCO mAP@5 (ViT-B/32) | mAP@5: 13.53 -> 31.32 | Improved |
| fixed-weight AVG fusion of T2I and I2I vs T2I only (ablation) | WISER · CIRCO mAP@5 (ViT-B/32) | mAP@5: 17.28 -> 13.53 | Hurt |
| add adaptive fusion over both paths (ablation) | WISER · Fashion-IQ val avg R@10 (ViT-B/32) | R@10: 33.40 -> 40.83 | Improved |
| add Deeper Thinking on top of adaptive fusion (ablation) | WISER · Fashion-IQ val avg R@10 (ViT-B/32) | R@10: 40.83 -> 41.99 | Improved |
| add Deeper Thinking on top of adaptive fusion (ablation) | WISER · CIRCO mAP@5 (ViT-B/32) | mAP@5: 31.32 -> 32.23 | Improved |
| add Deeper Thinking to the T2I-only path (ablation) | T2I baseline · Fashion-IQ val avg R@10 (ViT-B/32) | R@10: 28.59 -> 29.22 | Improved |
| add Wider Search (use both T2I and I2I with adaptive fusion) vs T2I alone (ablation) | WISER · CIRCO mAP@5 (ViT-B/32) | mAP@5: 17.28 -> 31.32 | Improved |
| use stronger editor model alone (BAGEL T2I) vs CIReVL (ablation) | T2I baseline · Fashion-IQ R@10 | R@10: 28.59 vs 28.27 | No clear effect |
| swap refiner LLM (Qwen-Turbo, GPT-3.5, GPT-4o-Mini, GPT-4o) (ablation) | WISER · CIRCO mAP@5 | mAP@5: 32.21 to 32.80 | No clear effect |
| scale verifier from 3B to 7B (ablation) | WISER · CIRCO mAP@5 | mAP@5: 27.50 -> 32.23 | Improved |
| scale verifier from 7B to 32B (ablation) | WISER · CIRCO mAP@5 | mAP@5: 32.23 -> 31.69 | Hurt |
| swap editor (OmniGen2, Step1X-Edit, Bagel-Edit vs Bagel) (ablation) | WISER · CIRCO mAP@5 | mAP@5: 31.18 to 32.23 | No clear effect |
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.
- CoTMR — CoTMR: Chain-of-Thought Multi-Scale Reasoning for Training-Free Zero-Shot Composed Image Retrieval (ICCV 2025) · better on 8 of 8 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 0.92)
- Instance-Level Composed Image Retrieval (NeurIPS 2025 · similarity 0.91)
- Context-I2W: Mapping Images to Context-Dependent Words for Accurate Zero-Shot Composed Image Retrieval (AAAI 2024 · similarity 0.90)
- Vision-by-Language for Training-Free Compositional Image Retrieval (ICLR 2024 · similarity 0.90)
- CoLLM: A Large Language Model for Composed Image Retrieval (CVPR 2025 · similarity 0.90)
- Language-only Training of Zero-shot Composed Image Retrieval (CVPR 2024 · similarity 0.90)
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.
