MoGe

MoGe is an advanced model for reconstructing accurate 3D geometry from a single image or video. With just a simple photo, MoGe can generate detailed 3D point maps and depth estimations, making it ideal for creating immersive visual content. Leveraging a ViT (Vision Transformer) encoder and a convolutional decoder, MoGe outputs high-quality depth maps, point maps, and 3D meshes. It also estimates complex properties like camera shift, focal length, and depth, providing a comprehensive view of spatial structure in images.

Key Features:

• Monocular 3D Reconstruction: Turns single images into accurate 3D point maps and meshes, even with challenging open-domain images.

• Support for Various Image Resolutions: Capable of handling a wide range of resolutions and aspect ratios (2:1 to 1:2).

• Fast Inference: Generates results in under 0.2 seconds on GPUs (A100 or RTX 3090).

• High-Quality Depth Range: Supports depth estimations for near and far distances with a range up to 1000x.

• Interactive Demos Available: Explore MoGe’s results on our Hugging Face demo page.