Improving fine-grained understanding in image-text pre-training

TL;DR

SPARC is a novel pretraining method that enhances fine-grained multimodal representations by learning sparse, token-specific image patch groupings, improving performance on both coarse and fine-grained vision-language tasks.

Contribution

The paper introduces SPARC, a simple yet effective approach for fine-grained contrastive alignment that learns sparse, token-specific image patch groupings without requiring batch negatives.

Findings

Improves performance on image classification, retrieval, detection, and segmentation tasks.

Enhances model faithfulness and captioning quality.

Achieves better fine-grained understanding compared to existing methods.

Abstract

We introduce SPARse Fine-grained Contrastive Alignment (SPARC), a simple method for pretraining more fine-grained multimodal representations from image-text pairs. Given that multiple image patches often correspond to single words, we propose to learn a grouping of image patches for every token in the caption. To achieve this, we use a sparse similarity metric between image patches and language tokens and compute for each token a language-grouped vision embedding as the weighted average of patches. The token and language-grouped vision embeddings are then contrasted through a fine-grained sequence-wise loss that only depends on individual samples and does not require other batch samples as negatives. This enables more detailed information to be learned in a computationally inexpensive manner. SPARC combines this fine-grained loss with a contrastive loss between global image and text embeddings to learn representations that simultaneously encode global and local information. We thoroughly evaluate our proposed method and show improved performance over competing approaches both on image-level tasks relying on coarse-grained information, e.g. classification, as well as region-level tasks relying on fine-grained information, e.g. retrieval, object detection, and segmentation. Moreover, SPARC improves model faithfulness and captioning in foundational vision-language models.