Ask or search anything...

TripoSR, developed by Stability AI and Tripo AI, reconstructs high-quality 3D meshes from a single 2D image in approximately 0.5 seconds. It achieves competitive performance in both reconstruction quality and inference speed by building upon the LRM architecture with several key innovations.

Long-LRM++ introduces a framework for real-time, feed-forward wide-coverage 3D reconstruction by combining semi-explicit feature Gaussians with a lightweight neural decoder. This approach achieves rendering quality comparable to state-of-the-art implicit methods while maintaining real-time frame rates of 14 FPS, representing an 8x speedup over prior high-fidelity models, and effectively preserves fine details.