LSGQuant: Layer-Sensitivity Guided Quantization for One-Step Diffusion Real-World Video Super-Resolution

2026-02-03

AI Summary

This work addresses the challenges of deploying one-step diffusion models for real-world video super-resolution, where their large model size and high computational cost hinder practical application, and aggressive low-bit quantization suffers from performance degradation due to wide activation dynamic ranges and heterogeneous layer behaviors. To this end, the authors propose a layer sensitivity-guided quantization framework, introducing layer sensitivity analysis into diffusion model quantization for the first time. The approach integrates a Dynamic Range-Adaptive Quantizer (DRAQ), a Variance-Oriented Layer Training Strategy (VOLTS), and Quantization-Aware Joint Optimization (QAO), leveraging layer-wise statistics collected during calibration for fine-grained quantization. Experimental results demonstrate that the proposed method nearly preserves full-precision model performance even at extremely low bit widths, significantly outperforming existing quantization techniques.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Calibration & Uncertainty Quantification
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Large language models for search
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

One-Step Diffusion Models have demonstrated promising capability and fast inference in video super-resolution (VSR) for real-world. Nevertheless, the substantial model size and high computational cost of Diffusion Transformers (DiTs) limit downstream applications. While low-bit quantization is a common approach for model compression, the effectiveness of quantized models is challenged by the high dynamic range of input latent and diverse layer behaviors. To deal with these challenges, we introduce LSGQuant, a layer-sensitivity guided quantizing approach for one-step diffusion-based real-world VSR. Our method incorporates a Dynamic Range Adaptive Quantizer (DRAQ) to fit video token activations. Furthermore, we estimate layer sensitivity and implement a Variance-Oriented Layer Training Strategy (VOLTS) by analyzing layer-wise statistics in calibration. We also introduce Quantization-Aware Optimization (QAO) to jointly refine the quantized branch and a retained high-precision branch. Extensive experiments demonstrate that our method has nearly performance to origin model with full-precision and significantly exceeds existing quantization techniques. Code is available at: https://github.com/zhengchen1999/LSGQuant.

Problem

• video super-resolution
• one-step diffusion models
• model quantization
• diffusion transformers
• real-world VSR

Innovation

• Layer-Sensitivity Guided Quantization
• Dynamic Range Adaptive Quantizer
• Variance-Oriented Layer Training Strategy
• Quantization-Aware Optimization
• One-Step Diffusion Models