Luis A. Leiva

Training-Free Pseudo-Fusion for Composed Image Retrieval via Diffusion Models and Multimodal Large Language Models

This work proposes PeFuse (for pseudo-fusion), a training-free framework that leverages pretrained Diffusion Models and Multimodal Large Language Models to bridge modalities via generative conversion, and introduces two novel strategies: uni-directional and bi-directional conversion, which convert CIR into four single-...