Contrastive Flow Matching

2025-06-05

AI Summary

In conditional flow matching, ambiguous generation arises from flow path entanglement across distinct classes or text prompts. Method: This paper introduces Contrastive Flow Matching (CFM), the first framework to incorporate contrastive learning into flow matching, explicitly maximizing inter-condition divergence in predicted velocity fields to enhance conditional controllability—achieved without architectural modifications. Experiments unify modeling on ImageNet-1K (class-conditional) and CC3M (text-to-image). Results: CFM accelerates training by up to 9×, reduces sampling steps by 5×, and improves FID by up to 8.9, consistently outperforming standard flow matching baselines. Its core innovation lies in the explicit decoupling of conditional flows—overcoming the fundamental limitation of flow overlap in conventional conditional flow matching—thereby simultaneously advancing generation quality, inference efficiency, and conditional fidelity.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Natural Language Processing: Generation
• Machine Learning: Deep Generative Models & Autoencoders

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Unconditional flow-matching trains diffusion models to transport samples from a source distribution to a target distribution by enforcing that the flows between sample pairs are unique. However, in conditional settings (e.g., class-conditioned models), this uniqueness is no longer guaranteed--flows from different conditions may overlap, leading to more ambiguous generations. We introduce Contrastive Flow Matching, an extension to the flow matching objective that explicitly enforces uniqueness across all conditional flows, enhancing condition separation. Our approach adds a contrastive objective that maximizes dissimilarities between predicted flows from arbitrary sample pairs. We validate Contrastive Flow Matching by conducting extensive experiments across varying model architectures on both class-conditioned (ImageNet-1k) and text-to-image (CC3M) benchmarks. Notably, we find that training models with Contrastive Flow Matching (1) improves training speed by a factor of up to 9x, (2) requires up to 5x fewer de-noising steps and (3) lowers FID by up to 8.9 compared to training the same models with flow matching. We release our code at: https://github.com/gstoica27/DeltaFM.git.

Problem

• Enhances uniqueness in conditional flow matching for diffusion models
• Reduces flow ambiguity between different conditional generations
• Improves training speed and generation quality in conditional settings

Innovation

• Enforces uniqueness across conditional flows
• Adds contrastive objective for flow dissimilarities
• Improves training speed and reduces de-noising steps