PID: Physics-Informed Diffusion Model for Infrared Image Generation

2024-07-12

Pattern Recognition

AI Summary

Existing RGB-to-infrared (IR) translation methods typically treat IR synthesis as a style transfer task, neglecting the underlying physical imaging process—leading to domain shift and thermal radiation distortion. This work proposes the first diffusion-based framework explicitly incorporating infrared physical priors. Specifically, we embed Planck’s radiation law and atmospheric attenuation models directly into the denoising iterations of a Denoising Diffusion Probabilistic Model (DDPM). Through a physics-driven loss function and gradient-guided sampling, our approach jointly optimizes radiometric consistency and perceptual fidelity during both training and inference—without introducing additional learnable parameters. Evaluated on multiple benchmarks, our method achieves new state-of-the-art performance: +2.1 dB PSNR and +0.08 SSIM over prior arts. Critically, generated IR images exhibit verifiable thermal radiation distributions, establishing a physically grounded paradigm for credible IR synthesis under low-visibility conditions.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Planning, Routing, and Scheduling: Planning with Language Models
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Security and Privacy: Data transparency and provenance
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Infrared imaging technology has gained significant attention for its reliable sensing ability in low visibility conditions, prompting many studies to convert the abundant RGB images to infrared images. However, most existing image translation methods treat infrared images as a stylistic variation, neglecting the underlying physical laws, which limits their practical application. To address these issues, we propose a Physics-Informed Diffusion (PID) model for translating RGB images to infrared images that adhere to physical laws. Our method leverages the iterative optimization of the diffusion model and incorporates strong physical constraints based on prior knowledge of infrared laws during training. This approach enhances the similarity between translated infrared images and the real infrared domain without increasing extra training parameters. Experimental results demonstrate that PID significantly outperforms existing state-of-the-art methods. Our code is available at https://github.com/fangyuanmao/PID.

Problem

• Converts RGB images to physically accurate infrared images
• Incorporates physical laws into diffusion model training
• Enhances infrared image realism without extra parameters

Innovation

• Physics-Informed Diffusion model for infrared generation
• Incorporates physical constraints during training
• Enhances similarity without extra parameters