Low-Rank Similarity Mining for Multimodal Dataset Distillation

TL;DR

This paper introduces Low-Rank Similarity Mining (LoRS), a novel method for distilling multimodal datasets like image-text pairs by efficiently capturing similarity matrices, improving scalability and performance in visual-language dataset distillation.

Contribution

LoRS is the first approach to simultaneously distill similarity matrices and utilize low-rank factorization for multimodal dataset distillation, enhancing efficiency and effectiveness.

Findings

Significant improvement over existing algorithms.

Efficient scalability through low-rank factorization.

Effective in visual-language dataset distillation.

Abstract

Though dataset distillation has witnessed rapid development in recent years, the distillation of multimodal data, e.g., image-text pairs, poses unique and under-explored challenges. Unlike unimodal data, image-text contrastive learning (ITC) data lack inherent categorization and should instead place greater emphasis on modality correspondence. In this work, we propose Low-Rank Similarity Mining (LoRS) for multimodal dataset distillation, that concurrently distills a ground truth similarity matrix with image-text pairs, and leverages low-rank factorization for efficiency and scalability. The proposed approach brings significant improvement to the existing algorithms, marking a significant contribution to the field of visual-language dataset distillation. We advocate adopting LoRS as a foundational synthetic data setup for image-text dataset distillation. Our code is available at https://github.com/silicx/LoRS_Distill.