LGCVOct 24, 2024

Stable Consistency Tuning: Understanding and Improving Consistency Models

arXiv:2410.18958v310 citationsh-index: 13Has Code
Originality Highly original
AI Analysis

This work addresses the efficiency problem in generative AI for applications requiring fast image synthesis, though it is incremental as it builds upon existing consistency tuning methods.

The paper tackles the slow generation speed of diffusion models by improving consistency models, achieving state-of-the-art results with 1-step FID 2.42 and 2-step FID 1.55 on ImageNet-64.

Diffusion models achieve superior generation quality but suffer from slow generation speed due to the iterative nature of denoising. In contrast, consistency models, a new generative family, achieve competitive performance with significantly faster sampling. These models are trained either through consistency distillation, which leverages pretrained diffusion models, or consistency training/tuning directly from raw data. In this work, we propose a novel framework for understanding consistency models by modeling the denoising process of the diffusion model as a Markov Decision Process (MDP) and framing consistency model training as the value estimation through Temporal Difference~(TD) Learning. More importantly, this framework allows us to analyze the limitations of current consistency training/tuning strategies. Built upon Easy Consistency Tuning (ECT), we propose Stable Consistency Tuning (SCT), which incorporates variance-reduced learning using the score identity. SCT leads to significant performance improvements on benchmarks such as CIFAR-10 and ImageNet-64. On ImageNet-64, SCT achieves 1-step FID 2.42 and 2-step FID 1.55, a new SoTA for consistency models.

Code Implementations1 repo
Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes