Refining Few-Step Text-to-Multiview Diffusion via Reinforcement Learning

2025-05-26

AI Summary

To address the degradation of image fidelity and cross-view consistency in few-step text-to-multi-view (T2MV) diffusion models, this paper proposes the first end-to-end reinforcement learning (RL)-based fine-tuning framework. We formulate multi-view denoising as a unified Markov decision process and introduce two key innovations: (i) ZMV-Sampling, a two-stage sampling strategy that coordinates view generation, and (ii) MV-ZigAL, a constraint mechanism enforcing geometric and semantic alignment across views. Crucially, we design the first joint geometry-semantic reward function to guide RL optimization. Our method achieves significant improvements in both single-view image quality and inter-view consistency using only a small number of denoising steps. It attains state-of-the-art performance across multiple T2MV benchmarks while maintaining high inference efficiency and generation fidelity.

Technology Category

• Machine Learning: Multi-instance/Multi-view Learning
• Computer Vision: Diffusion Models for Vision
• Natural Language Processing: Language Grounding & Multi-modal NLP

Application Category

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

Abstract

Text-to-multiview (T2MV) generation, which produces coherent multiview images from a single text prompt, remains computationally intensive, while accelerated T2MV methods using few-step diffusion models often sacrifice image fidelity and view consistency. To address this, we propose a novel reinforcement learning (RL) finetuning framework tailored for few-step T2MV diffusion models to jointly optimize per-view fidelity and cross-view consistency. Specifically, we first reformulate T2MV denoising across all views as a single unified Markov decision process, enabling multiview-aware policy optimization driven by a joint-view reward objective. Next, we introduce ZMV-Sampling, a test-time T2MV sampling technique that adds an inversion-denoising pass to reinforce both viewpoint and text conditioning, resulting in improved T2MV generation at the cost of inference time. To internalize its performance gains into the base sampling policy, we develop MV-ZigAL, a novel policy optimization strategy that uses reward advantages of ZMV-Sampling over standard sampling as learning signals for policy updates. Finally, noting that the joint-view reward objective under-optimizes per-view fidelity but naively optimizing single-view metrics neglects cross-view alignment, we reframe RL finetuning for T2MV diffusion models as a constrained optimization problem that maximizes per-view fidelity subject to an explicit joint-view constraint, thereby enabling more efficient and balanced policy updates. By integrating this constrained optimization paradigm with MV-ZigAL, we establish our complete RL finetuning framework, referred to as MVC-ZigAL, which effectively refines the few-step T2MV diffusion baseline in both fidelity and consistency while preserving its few-step efficiency.

Problem

• Optimizing few-step T2MV diffusion for fidelity and view consistency
• Developing RL framework for multiview-aware policy optimization
• Balancing per-view fidelity with cross-view alignment constraints

Innovation

• RL finetuning for multiview diffusion models
• ZMV-Sampling enhances viewpoint and text conditioning
• Constrained optimization balances fidelity and consistency