thu-ml/CRM
A feed-forward model that generates 3D textured meshes from a single image in 10 seconds using convolutional reconstruction and diffusion models.
Not currently ranked — collecting fresh signals.
star history
CRM (Convolutional Reconstruction Model) is a deep learning system that reconstructs complete 3D textured meshes from a single 2D image input. It leverages diffusion-based multiview generation and a reconstruction pipeline to produce usable 3D geometry with textures in under 10 seconds. The model is implemented in PyTorch with components like xformers and nvdiffrast for efficient rendering.
Frequently asked
- What is thu-ml/CRM?
- A feed-forward model that generates 3D textured meshes from a single image in 10 seconds using convolutional reconstruction and diffusion models.
- Is CRM open source?
- Yes — thu-ml/CRM is open source, released under the MIT license.
- What language is CRM written in?
- thu-ml/CRM is primarily written in Python.
- How popular is CRM?
- thu-ml/CRM has 690 stars on GitHub.
- Where can I find CRM?
- thu-ml/CRM is on GitHub at https://github.com/thu-ml/CRM.