Saved in:
Bibliographic Details
Main Authors: Xu, Yue, Lin, Zhilin, Qiu, Yusong, Lu, Cewu, Li, Yong-Lu
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2406.03793
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911907976314880
author Xu, Yue
Lin, Zhilin
Qiu, Yusong
Lu, Cewu
Li, Yong-Lu
author_facet Xu, Yue
Lin, Zhilin
Qiu, Yusong
Lu, Cewu
Li, Yong-Lu
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2406_03793
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Low-Rank Similarity Mining for Multimodal Dataset Distillation
Xu, Yue
Lin, Zhilin
Qiu, Yusong
Lu, Cewu
Li, Yong-Lu
Machine Learning
Computer Vision and Pattern Recognition
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.
title Low-Rank Similarity Mining for Multimodal Dataset Distillation
topic Machine Learning
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2406.03793