TripoSR
VAST-AI-Research/TripoSR · Python · MIT
Fast feed-forward image-to-3D
この解説はまだ日本語がありません。英語の原文を表示しています。
TripoSR prioritizes speed: drop an image, get a mesh quickly for blocking out game assets.
スコア
導入の要点
pip install triposr 深読み
TripoSR is a fast single-image-to-3D reconstruction model from VAST AI Research (with Stability AI collaboration): one photo in, a textured 3D model out in under a second on a capable GPU.
It builds on the LRM (Large Reconstruction Model) line of research — feed-forward 3D reconstruction without per-object optimization — which is what makes it fast enough for interactive use.
The repo is a compact, pip-installable inference stack with clear requirements (Python 3.8+, CUDA, PyTorch), so it drops into existing 3D tooling; the approach is documented in an arXiv paper.