VAST-AI-Research/TripoSR
TripoSR
TripoSR is a single-image 3D reconstruction system that generates a high-quality textured mesh from one photograph in under half a second. It uses a feedforward neural network to process a single image through a transformer architecture, compressing the input into a compact latent vector that conditions the entire reconstruction pipeline. The system outputs a separate UV texture map with configurable resolution, replacing vertex colors for higher-quality surface detail.
The project is built around an end-to-end differentiable pipeline that trains the entire reconstruction system from image input to 3D output using gradient-based optimization without intermediate steps. It encodes 3D shape and appearance as three orthogonal 2D feature planes that are decoded into a textured mesh, enabling sub-second feedforward inference by avoiding iterative optimization or test-time refinement. The texture baking pipeline produces a separate texture map decoupled from the vertex-based geometry output, with adjustable resolution for enhanced surface detail.
The system covers rapid single-image 3D asset creation, feedforward neural network inference for geometry, and texture extraction with configurable resolution. It produces ready-to-use 3D assets from minimal visual input, running a neural network that reconstructs 3D geometry in a single forward pass.
Features
- Single-Image 3D Reconstructions - Generates a high-quality 3D model from a single photograph in under half a second using a feedforward neural network.
- Triplane Representations - Encodes 3D shape and appearance as three orthogonal 2D feature planes that are decoded into a textured mesh.
- End-to-End Architectures - Trains the entire reconstruction system from image input to 3D output using gradient-based optimization without intermediate steps.
- Latent Space Encoders - Compresses the input image into a compact latent vector that conditions the entire 3D reconstruction pipeline.
- 3D Reconstruction Transformers - Processes a single image through a transformer network to directly predict 3D geometry and texture in one forward pass.
- Texture Map Baking - Outputs a separate UV texture map with configurable resolution, replacing vertex colors for higher-quality surface detail.
- UV Texture Map Generators - Produces a separate UV texture map with configurable resolution, decoupled from the vertex-based geometry output.
- 3D Asset Generators - Produces ready-to-use 3D assets for games, VR, or design from minimal visual input.
- Sub-Second Feedforward Predictors - Generates a complete 3D model in under 500 milliseconds by avoiding iterative optimization or test-time refinement.
- Real-Time 3D Object Generators - Produces a complete 3D model from one image in under half a second, enabling interactive or batch workflows.
- Neural Rendering - Listed in the “Neural Rendering” section of the Ailia Models awesome list.
