[CVPR 2026] WISER: Wider Search, Deeper Thinking, and Adaptive Fusion for Training-Free Zero-Shot Composed Image Retrieval

If you find this project useful, please give us a star 🌟.

🔥 News

• [2026/2/23] Our paper has been accepted to CVPR 2026!
• [2026/2/27] We release our paper in the arxiv.
• [2026/2/27] We release the code.

🔎 Overview

Overview of the proposed WISER framework. (1) Wider Search. We leverage an editor to produce text and image queries for dual-path retrieval, aggregating the top-K results into a unified candidate pool. (2) Adaptive Fusion. We employ a verifier to assess the candidates with confidence scores, applying a multi-level fusion strategy for high-confidence results and triggering refinement for low-confidence ones. (3) Deeper Thinking. For uncertain retrievals, we leverage a refiner to analyze unmet modifications and then feed targeted suggestions back to the editor, iterating until a predefined limit is reached.

⚡️ Getting Started

Requirements

conda create -n wiser python=3.10
conda activate wiser
pip install -r requirements.txt

Data Preparing

Follow the dataset preparation from CIReVL. After downloading and organizing the datasets, update paths and parameters in ./config/start_config_circo.json.

How to Run

Here we take CIRCO dataset as an example.

Step1: Prepare Gallery Captions

bash run_step1.sh

Step 2: Inference

bash run_step2.sh

📊 Main Results

WISER significantly outperforms previous methods across multiple benchmarks, achieving relative improvements of 45% on CIRCO (mAP@5) and 57% on CIRR (Recall@1) over existing training-free methods. Notably, it even surpasses many training-dependent methods, highlighting its superiority and generalization under diverse scenarios.

    

📄 Citation

@article{wang2026wiser,
  title={WISER: Wider Search, Deeper Thinking, and Adaptive Fusion for Training-Free Zero-Shot Composed Image Retrieval},
  author={Wang, Tianyue and Qu, Leigang and Yang, Tianyu and Hao, Xiangzhao and Xu, Yifan and Guo, Haiyun and Wang, Jinqiao},
  journal={arXiv preprint arXiv:2602.23029},
  year={2026}
}