Multi-View Photometric Stereo

This repo contains the photometric normal refinement routines.

For geometric coarse normal estimation, see this repo.

Dependencies

python>=3.7 numpy skimage matplotlib scipy python-opencv open3d==0.16.0

File structure

Implemented in stereo_flash_no_flash.py, evaluate.py, pointcloud_processing.py.

Blender simulation and exporting scripts in blender/.

Configuration is in configs.py.

Helper functions are in utils.py, cp_hw2.py.

Place the data in datasets/.

How to run

To perform reconstruction, and run the following steps:

Step 0: Change config

Set the object name in configs.py. Currently supported: bunny, buddha, shoe.

Step 1: Process point cloud

Turn point cloud from SfM toolbox into a mesh and then into a coarse normal map.

python3 pointcloud_processing.py

Step 2: Evaluate normal refinement

python3 evaluate.py