Timestep-Aware Correction for Quantized Diffusion Models

TL;DR

This paper introduces a timestep-aware correction method for quantized diffusion models that dynamically reduces quantization errors during image generation, significantly improving output quality on resource-limited devices.

Contribution

The paper presents a novel timestep-aware correction technique that addresses error accumulation in low-precision diffusion models, enhancing their performance with minimal additional computation.

Findings

Achieves state-of-the-art results on low-precision diffusion models.

Substantially improves image quality with negligible computational overhead.

Demonstrates generalizability across different diffusion model architectures.

Abstract

Diffusion models have marked a significant breakthrough in the synthesis of semantically coherent images. However, their extensive noise estimation networks and the iterative generation process limit their wider application, particularly on resource-constrained platforms like mobile devices. Existing post-training quantization (PTQ) methods have managed to compress diffusion models to low precision. Nevertheless, due to the iterative nature of diffusion models, quantization errors tend to accumulate throughout the generation process. This accumulation of error becomes particularly problematic in low-precision scenarios, leading to significant distortions in the generated images. We attribute this accumulation issue to two main causes: error propagation and exposure bias. To address these problems, we propose a timestep-aware correction method for quantized diffusion model, which dynamically corrects the quantization error. By leveraging the proposed method in low-precision diffusion models, substantial enhancement of output quality could be achieved with only negligible computation overhead. Extensive experiments underscore our method's effectiveness and generalizability. By employing the proposed correction strategy, we achieve state-of-the-art (SOTA) results on low-precision models.