Application of denoising diffusion probabilistic model in the Minecraft environment

In this paper, we study the influence of hyperparameters of image generation model in Denoising Diffusion Probabilistic Model (DDPM) on image generation. Experiments were performed on MineRL dataset, the batch size, dimensions, learning rate, and sampling time steps of the model were adjusted, and the Fréchet Inception Distance (FID) was used to evaluate the quality of the resulting images. We introduce a performance degradation index to compare the effects of different hyperparameter settings. The experimental results show that dimension and learning rate are the key factors affecting the quality of DDPM image generation. These findings are of great significance for optimizing the DDPM model and improving its performance in image generation tasks.