Tripo S R

Introduction

TripoSR is a fast, feed-forward 3D generative model developed collaboratively by Stability AI and Tripo AI. The model leverages advanced techniques to generate 3D reconstructions from single images efficiently.

Architecture

TripoSR follows the LRM network architecture, incorporating significant advancements in data curation and model training. These improvements enhance the model's performance in generating realistic 3D models. For detailed technical insights, refer to the provided tech report.

• Collaborators: Stability AI, Tripo AI
• Model Type: Feed-forward 3D reconstruction from a single image
• Hardware: Trained over 5 days on 22 GPU nodes, each with 8 A100 40GB GPUs

Training

The model was trained using a subset of the Objaverse dataset, utilizing enhanced rendering methods to better replicate real-world image distributions. This approach aids in improving the model's generalization capabilities. The dataset is available under the CC-BY license.

Guide: Running Locally

To run TripoSR locally, follow these steps:

1. Clone the Repository:

   git clone https://github.com/VAST-AI-Research/TripoSR
   cd TripoSR
   

2. Install Dependencies: Make sure to install all the necessary Python packages as listed in the requirements.txt.

3. Run the Model: Follow the instructions in the repository's README to execute the model on your local setup.

4. Cloud GPUs: For optimal performance, consider using cloud-based GPUs such as AWS EC2 P3 instances, Google Cloud TPUs, or Azure's NDv4 series.

License

TripoSR is licensed under the MIT License, which permits reuse with proper attribution. For more information, refer to the license details in the repository.