ITO: Images and Texts as One via Synergizing Multiple Alignment and Training-Time Fusion

TL;DR

ITO introduces a novel framework that combines multiple alignment and training-time fusion to improve image-text representations, effectively reducing modality gaps and enhancing performance across various benchmarks.

Contribution

The paper presents ITO, a new method that synergizes multimodal alignment with a lightweight fusion module, improving cross-modal learning without increasing inference complexity.

Findings

Outperforms strong baselines in classification, retrieval, and multimodal tasks.

Training-time fusion acts as a structural regularizer, stabilizing training.

Multiple alignment enhances discriminative power of representations.

Abstract

Image-text contrastive pretraining has become a dominant paradigm for visual representation learning, yet existing methods often yield representations that remain partially organized by modality. We propose ITO, a framework addressing this limitation through two synergistic mechanisms. Multimodal multiple alignment enriches supervision by mining diverse image-text correspondences, while a lightweight training-time multimodal fusion module enforces structured cross-modal interaction. Crucially, the fusion module is discarded at inference, preserving the efficiency of standard dual-encoder architectures. Extensive experiments show that ITO consistently outperforms strong baselines across classification, retrieval, and multimodal benchmarks. Our analysis reveals that while multiple alignment drives discriminative power, training-time fusion acts as a critical structural regularizer -- eliminating the modality gap and stabilizing training dynamics to prevent the early saturation often observed in aggressive contrastive learning.