Monocular Depth & Geometric Estimation

12 repos

Zero-shot depth and surface normal estimation from single images using diffusion models, enabling 3D scene understanding from monocular video and photos without task-specific training data. The cluster centers on the Marigold family of models, which leverage pre-trained diffusion foundations for in-the-wild geometric prediction. Repositories here span model variants (depth, normals, lighting estimation), inference implementations, and related diffusion-based vision approaches that extend beyond depth to other 3D geometric tasks.

Related clusters