Siddharth Dey

Depth Latent Diffusion

Trained a conditional Latent-Diffusion Model for monocular depth estimation on the NYU dataset.

You can find the project report here [Link]

Results

NYU_progrssing_denoising

Progressive denoising of the latent space conditioned on the RGB image to generate the depth map

NYU_progrssing_denoising

Result comparison between the spatial rescaler and using DINOv2 for RBG conditioning