TiMix: Text-Aware Image Mixing for Effective Vision-Language Pre-training
Chaoya Jiang, Wei Ye, Haiyang Xu, Qinghao Ye, Ming Yan, Ji Zhang, Shikun Zhang
Abstract
Self-supervised Multi-modal Contrastive Learning (SMCL) remarkably advances modern Vision-Language Pre-training (VLP) models by aligning visual and linguistic modalities. Due to noises in web-harvested text-image pairs, however, scaling up training data volume in SMCL presents considerable obstacles in terms of computational cost and data inefficiency. To improve data efficiency in VLP, we propose Text-aware Image Mixing (TiMix), which integrates mix-based data augmentation techniques into SMCL, yielding significant performance improvements without significantly increasing computational overhead. We provide a theoretical analysis of TiMix from a mutual information (MI) perspective, showing that mixed data samples for cross-modal contrastive learning implicitly serve as a regularizer for the contrastive loss. The experimental results demonstrate that TiMix exhibits a comparable performance on downstream tasks, even with a reduced amount of training data and shorter training time, when benchmarked against existing methods. This work empirically and theoretically demonstrates the potential of data mixing for data-efficient and computationally viable VLP, benefiting broader VLP model adoption in practical scenarios. Our code is available on https://github.com/chaoyajiang/TiMiX/tree/main.
BibTeX
@article{Jiang_Ye_Xu_Ye_Yan_Zhang_Zhang_2024, title={TiMix: Text-Aware Image Mixing for Effective Vision-Language Pre-training}, volume={38}, url={https://ojs.aaai.org/index.php/AAAI/article/view/28025}, DOI={10.1609/aaai.v38i3.28025}, abstractNote={Self-supervised Multi-modal Contrastive Learning (SMCL) remarkably advances modern Vision-Language Pre-training (VLP) models by aligning visual and linguistic modalities. Due to noises in web-harvested text-image pairs, however, scaling up training data volume in SMCL presents considerable obstacles in terms of computational cost and data inefficiency. To improve data efficiency in VLP, we propose Text-aware Image Mixing (TiMix), which integrates mix-based data augmentation techniques into SMCL, yielding significant performance improvements without significantly increasing computational overhead. We provide a theoretical analysis of TiMix from a mutual information (MI) perspective, showing that mixed data samples for cross-modal contrastive learning implicitly serve as a regularizer for the contrastive loss. The experimental results demonstrate that TiMix exhibits a comparable performance on downstream tasks, even with a reduced amount of training data and shorter training time, when benchmarked against existing methods. This work empirically and theoretically demonstrates the potential of data mixing for data-efficient and computationally viable VLP, benefiting broader VLP model adoption in practical scenarios. Our code is available on https://github.com/chaoyajiang/TiMiX/tree/main.}, number={3}, journal={Proceedings of the AAAI Conference on Artificial Intelligence}, author={Jiang, Chaoya and Ye, Wei and Xu, Haiyang and Ye, Qinghao and Yan, Ming and Zhang, Ji and Zhang, Shikun}, year={2024}, month={Mar.}, pages={2489-2497} }