Depth To Normal

Python code used to transform depth map to normal map using point cloud as intermediate representation

Compilation

This repo uses the code from "Fast and robust normal estimation for point clouds with sharp features" by Alexandre Boulch and Renaud Marlet to estimate the normal map from point clouds.

python wrapper

Using the library with Python requires building the wrapper. This works fine with g++ 5.4.0 on Ubuntu 16.04 with Anaconda.

cd python
python setup.py install --home="."

python test

python example.py

The script creates a point cloud on a cube and estimate the normals. It produces a .xyz containing both points and normals, it can be displayed using Meshlab.

Usage

Our usage is based on the dataset providing accurate RGB-D image pairs such as iBims-1

First, use depth_to_pointcloud.py to generate point clouds from depth maps.

• the default format for depth map is in uint16

• some pixels in depth map is labeled as 0 as a result of invalid depth estimation

• the generated point cloud is of format [x, y, z] in meters

Then, use pointcloud_to_normal.py to generate normal maps from point clouds.

• here we provide also a function to generate point to point depth map from plane to plane depth map

• the normal_xyz output saves the point cloud with normal map in .xyz file of format [x, y, z, n_x, n_y, n_z]

• the normal_img output saves the normal map in .png file with RGB values corresponding to normal vectors

License

See here for the original license.