Diff-ICMH: Harmonizing Machine and Human Vision in Image Compression with Generative Prior

2025-11-27

AI Summary

Existing image compression methods typically optimize for either human perception or machine analysis in isolation, failing to jointly satisfy both objectives. This paper proposes Diff-ICMH—the first generative image compression framework explicitly designed to co-optimize for both human visual quality and machine vision utility. Our key contributions are threefold: (1) We identify shared principles between semantic fidelity and perceptual realism across vision tasks, enabling the design of a semantic consistency loss (SC loss) and a label-guided module (TGM) that leverages image-level semantic labels to enhance semantic reconstruction; (2) We construct a unified encoder-decoder architecture grounded in diffusion priors, supporting multiple downstream AI tasks and high-fidelity subjective quality from a single bitstream and a single codec; (3) Extensive experiments demonstrate state-of-the-art performance—superior generalization on classification and detection tasks, alongside top-tier PSNR, MS-SSIM, and subjective quality scores.

Technology Category

Computer Vision: Diffusion Models for VisionMachine Learning: Deep Generative Models & AutoencodersHumans and AI: Crowd Sourcing and Human Computation

Application Category

Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labelingSemantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMsUser Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Image compression methods are usually optimized isolatedly for human perception or machine analysis tasks. We reveal fundamental commonalities between these objectives: preserving accurate semantic information is paramount, as it directly dictates the integrity of critical information for intelligent tasks and aids human understanding. Concurrently, enhanced perceptual quality not only improves visual appeal but also, by ensuring realistic image distributions, benefits semantic feature extraction for machine tasks. Based on this insight, we propose Diff-ICMH, a generative image compression framework aiming for harmonizing machine and human vision in image compression. It ensures perceptual realism by leveraging generative priors and simultaneously guarantees semantic fidelity through the incorporation of Semantic Consistency loss (SC loss) during training. Additionally, we introduce the Tag Guidance Module (TGM) that leverages highly semantic image-level tags to stimulate the pre-trained diffusion model's generative capabilities, requiring minimal additional bit rates. Consequently, Diff-ICMH supports multiple intelligent tasks through a single codec and bitstream without any task-specific adaptation, while preserving high-quality visual experience for human perception. Extensive experimental results demonstrate Diff-ICMH's superiority and generalizability across diverse tasks, while maintaining visual appeal for human perception. Code is available at: https://github.com/RuoyuFeng/Diff-ICMH.

Problem

• Harmonizing machine analysis and human visual perception in image compression
• Optimizing compression for both semantic fidelity and perceptual quality simultaneously
• Enabling single codec support for multiple intelligent tasks without adaptation

Innovation

• Leverages generative priors for perceptual realism
• Uses Semantic Consistency loss for semantic fidelity
• Introduces Tag Guidance Module for enhanced generative capabilities