Modeling Caption Diversity in Contrastive Vision-Language Pretraining

TL;DR

Llip introduces a novel approach to model the diversity of image captions in contrastive vision-language pretraining, leading to improved zero-shot classification and retrieval performance over traditional CLIP models.

Contribution

Llip models caption diversity by integrating multiple visual features conditioned on text, enhancing representation richness beyond single-vector methods like CLIP.

Findings

Llip outperforms CLIP and SigLIP on various tasks.

Achieves 83.5% top-1 accuracy on ImageNet zero-shot classification.

Improves zero-shot retrieval on MS-COCO by 6.0%.

Abstract

There are a thousand ways to caption an image. Contrastive Language Pretraining (CLIP) on the other hand, works by mapping an image and its caption to a single vector -- limiting how well CLIP-like models can represent the diverse ways to describe an image. In this work, we introduce Llip, Latent Language Image Pretraining, which models the diversity of captions that could match an image. Llip's vision encoder outputs a set of visual features that are mixed into a final representation by conditioning on information derived from the text. We show that Llip outperforms non-contextualized baselines like CLIP and SigLIP on a variety of tasks even with large-scale encoders. Llip improves zero-shot classification by an average of 2.9% zero-shot classification benchmarks with a ViT-G/14 encoder. Specifically, Llip attains a zero-shot top-1 accuracy of 83.5% on ImageNet outperforming a similarly sized CLIP by 1.4%. We also demonstrate improvement on zero-shot retrieval on MS-COCO by 6.0%. We provide a comprehensive analysis of the components introduced by the method and demonstrate that Llip leads to richer visual representations.