DAG: Depth-Aware Guidance with Denoising Diffusion Probabilistic Models

RGB-D-Fusion: Image Conditioned Depth Diffusion of Humanoid Subjects

“…Not as closely related but still applying diffusion models on 3D related data representations, we found 3D point cloud generation conditioned on monocular images [75], conditioned on an encoded shape latent [76] and conditioned in CLIP-tokens [77], novel view synthesis from a single view [78], [79], for perpetual view generations for long camera trajectories where depth is predicted as an intermediate representation [80] or combining text prompts for 2D generation with Neural Radiance Fields (NeRFs) [81], depth estimation from multiple camera images at different viewpoints [82] and depth-aware guidance methods that guide the image generation process by its intermediate depth representation [83].…”