TripoSF: High-Resolution 3D Shape Modeling with SparseFlex

TripoSF is a state-of-the-art 3D shape modeling framework that enables differentiable mesh reconstruction at resolutions up to $1024^3$ directly from rendering losses. This repository contains the pretrained VAE model for high-fidelity 3D reconstruction.

Model Description

TripoSF leverages a novel SparseFlex representation that combines the accuracy of Flexicubes with an efficient sparse voxel structure, focusing computation on surface-adjacent regions.

Key Features

• 🔍 Ultra-high resolution reconstruction (up to $1024^3$)
• 🎯 Direct optimization from rendering losses
• 🌐 Natural handling of open surfaces and complex topologies
• 💾 Memory-efficient sparse computation
• 🔄 Differentiable mesh extraction with sharp features

Intended Uses

This model is designed for:

• High-fidelity 3D shape reconstruction
• Mesh generation and modeling
• 3D asset creation and optimization

Requirements

• CUDA-capable GPU (≥12GB VRAM recommended for $1024^3$ resolution)
• PyTorch 2.0+

Usage

For detailed usage instructions, please visit our GitHub repository.

About

TripoSF is developed by Tripo, VAST AI Research, pushing the boundaries of 3D Generative AI. For more information:

• Project Page
• Paper
• GitHub Repository