keyword

Multi-Text Training Loss

Multi-text training loss is a contrastive loss function used in vision-language pre-training that simultaneously aligns an individual visual input with multiple corresponding text descriptions. Rather than pairing an image with only a single caption or randomly selecting a single text augmentation per training iteration, this formulation treats multiple diverse textual descriptions, such as original captions and rewritten variants, as positive matches for the same image. By optimizing a multi-positive contrastive objective across the batch, multi-text training loss leverages richer linguistic diversity in vocabulary and syntactic phrasing, thereby mitigating text overfitting and improving cross-modal representation alignment and zero-shot transfer performance.

1 item

Improving CLIP Training with Language Rewrites

Lijie Fan, Dilip Krishnan, Phillip Isola, Dina Katabi, Yonglong Tian

Why you should read this

Introduces LaCLIP, a simple strategy that uses large language models to generate diverse text augmentations during contrastive pre-training, substantially increasing zero-shot transfer accuracy across vision-language benchmarks without incurring extra training overhead.

Contrastive Language-Image Pre-training (CLIP) stands as one of the most effective and scalable methods for training transferable vision models using paired image and text data. CLIP models are trained using contrastive loss, which typically relies on data augmentations to prevent overfitting and shortcuts. However, in the CLIP training paradigm, data augmentations are exclusively applied to image inputs, while language inputs remain unchanged throughout the entire training process, limiting the exposure of diverse texts to the same image. In this paper, we introduce Language augmented CLIP (LaCLIP), a simple yet highly effective approach to enhance CLIP training through language rewrites. Leveraging the in-context learning capability of large language models, we rewrite the text descriptions associated with each image. These rewritten texts exhibit diversity in sentence structure and vocabulary while preserving the original key concepts and meanings. During training, LaCLIP randomly selects either the original texts or the rewritten versions as text augmentations for each image. Extensive experiments on CC3M, CC12M, RedCaps and LAION-400M datasets show that CLIP pre-training with language rewrites significantly improves the transfer performance without computation or memory overhead during training. Specifically for ImageNet zero-shot accuracy, LaCLIP outperforms CLIP by 8.2% on CC12M and 2.4% on LAION-400M. Code is available at this https URL.

Source

https://arxiv.org/pdf/2305.20088

Added

2026-09-26