Guiding Diffusion-based Reconstruction with Contrastive Signals for Balanced Visual Representation
This work aims to improve the visual representation capacity of CLIP, which is a foundational model for many downstream vision-language tasks, benefiting researchers and practitioners using CLIP.
The paper addresses the limited understanding capacity of CLIP's visual encoder, specifically its Discriminative Ability (D-Ability) and Detail Perceptual Ability (P-Ability). They propose Diffusion Contrastive Reconstruction (DCR), which integrates contrastive signals from reconstructed images into the diffusion process, theoretically showing it jointly optimizes both abilities and validating its effectiveness across benchmarks and multi-modal large language models.
The limited understanding capacity of the visual encoder in Contrastive Language-Image Pre-training (CLIP) has become a key bottleneck for downstream performance. This capacity includes both Discriminative Ability (D-Ability), which reflects class separability, and Detail Perceptual Ability (P-Ability), which focuses on fine-grained visual cues. Recent solutions use diffusion models to enhance representations by conditioning image reconstruction on CLIP visual tokens. We argue that such paradigms may compromise D-Ability and therefore fail to effectively address CLIP's representation limitations. To address this, we integrate contrastive signals into diffusion-based reconstruction to pursue more comprehensive visual representations. We begin with a straightforward design that augments the diffusion process with contrastive learning on input images. However, empirical results show that the naive combination suffers from gradient conflict and yields suboptimal performance. To balance the optimization, we introduce the Diffusion Contrastive Reconstruction (DCR), which unifies the learning objective. The key idea is to inject contrastive signals derived from each reconstructed image, rather than from the original input, into the diffusion process. Our theoretical analysis shows that the DCR loss can jointly optimize D-Ability and P-Ability. Extensive experiments across various benchmarks and multi-modal large language models validate the effectiveness of our method. The code is available at https://github.com/boyuh/DCR.