Research map / Vision-language and multimodal / CLIP and zero-shot
Generating a Paracosm for Training-Free Zero-Shot Composed Image Retrieval
Tong Wang, Yunhan Zhao, Shu Kong
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 synthetic counterparts of database images (ablation) | Paracosm (CLIP ViT-B/32) · CIRR test | Recall@1: 27.93 -> 32.27 | Improved |
| add synthetic counterparts of database images (ablation) | Paracosm (CLIP ViT-B/32) · CIRCO test | mAP@5: 18.29 -> 26.10 | Improved |
| add modification text to the query feature (ablation) | Paracosm (CLIP ViT-B/32) · CIRR test | Recall@1: 24.60 -> 32.27 | Improved |
| add modification text to the query feature (ablation) | Paracosm (CLIP ViT-B/32) · CIRCO test | mAP@5: 16.57 -> 26.10 | Improved |
| add mental image to description-based query (ablation) | Paracosm with synthetic counterparts, no t_mod · CIRR test | Recall@1: 22.46 -> 24.60 | Improved |
| add mental image to description-based query (ablation) | Paracosm with synthetic counterparts, no t_mod · CIRCO test | mAP@5: 16.72 -> 16.57 | Hurt |
| edit the reference image instead of T2I generation for mental image (ablation) | Paracosm · CIRR test | Recall@1: 31.71 -> 32.27 | Improved |
| use LongCat-Image-Edit instead of Qwen-Image-Edit (ablation) | Paracosm · CIRR test | Recall@1: 32.27 -> 32.12 | No clear effect |
| use GPT-4o instead of Qwen2.5-VL for descriptions (ablation) | Paracosm · CIRR test | Recall@1: 32.27 -> 31.86 | No clear effect |
| add brief descriptions of database images (ablation) | Paracosm · Fashion IQ val | Recall@10: 26.14 -> 26.06 | No clear effect |
| add detailed descriptions of database images to matching (ablation) | Paracosm · CIRR test | Recall@1: 32.27 -> 28.33 | Hurt |
| replace CLIP with OpenCLIP ViT-B/32 backbone (ablation) | Paracosm · CIRCO test | mAP@5: 26.10 -> 31.31 | Improved |
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 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 0.92)
- Vision-by-Language for Training-Free Compositional Image Retrieval (ICLR 2024 · similarity 0.91)
- Generative Zero-Shot Composed Image Retrieval (CVPR 2025 · similarity 0.90)
- Instance-Level Composed Image Retrieval (NeurIPS 2025 · similarity 0.90)
- CoLLM: A Large Language Model for Composed Image Retrieval (CVPR 2025 · similarity 0.90)
- Visual Delta Generator with Large Multi-modal Models for Semi-supervised 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.
