Image-Conditional Diffusion Transformer for Underwater Image Enhancement
The episode discusses Xingyang Nie's paper, "Image-Conditional Diffusion Transformer for Underwater Image Enhancement" (ICDT). The hosts detail how ICDT replaces a U-Net with a transformer architecture in a diffusion model and uses conditional latent space conversion. Improvements include a hybrid loss function that allows for faster sampling and better covariance estimation, leading to state-of-the-art results and practical efficiency.