nianticlabs/simplerecon
A PyTorch implementation of a deep learning method for 3D scene reconstruction from multi-view images using cost volume estimation without 3D convolutions.
Velocity · 7d
+1.0
★ / day
Trend
→steady
star history
SimpleRecon is a research implementation for training and testing multi-view stereo depth estimation models. The method performs 3D reconstruction by processing 2D image features and cost volumes, avoiding expensive 3D convolutions for efficiency. The codebase provides training pipelines, evaluation tools on ScanNet dataset, and visualization utilities for the ECCV 2022 paper.