Guiding Diffusion-based Reconstruction with Contrastive Signals for Balanced Visual Representation

TL;DR

This paper proposes a novel diffusion contrastive reconstruction method that integrates contrastive signals into diffusion-based image reconstruction to enhance the visual representation capacity of CLIP, balancing discriminative and detail perceptual abilities.

Contribution

It introduces DCR, a unified framework that combines contrastive learning with diffusion models to improve CLIP's visual representations, addressing its limitations in class separability and fine-grained detail perception.

Findings

DCR improves downstream task performance across benchmarks.

Contrastive signals from reconstructed images enhance representation quality.

Theoretical analysis confirms joint optimization of D-Ability and P-Ability.

Abstract

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.