This paper proposes the Heat Diffusion Model (HDM), a novel diffusion probabilistic model that generates more realistic images by considering the relationships between pixels. While existing denoising diffusion probabilistic models (DDPMs) process the entire image, HDM incorporates an attention mechanism between pixels, exploiting the fact that adjacent pixels are more likely to belong to the same object. By incorporating the discrete form of the two-dimensional heat equation into the diffusion and generative formulas of DDPM, HDM computes the relationships between adjacent pixels during image processing. Experimental results show that HDM generates higher-quality samples than existing models such as DDPM, consistent diffusion model (CDM), latent diffusion model (LDM), and vector quantized generative adversarial network (VQGAN).
