Simple python rasterizer implemented by OpenGL and C++

Key features:

Support both PINHOLE and camera models with distortion parameters like (OPENCV, OPENCV_FISHEYE)
Headless OpenGL rendering
Output numpy.ndarray directly

Requirements

cmake >= 3.16

Install

git clone --recursive https://github.com/liu115/renderpy    # clone with submodules
conda create -n renderpy python=3.9
conda activate renderpy
pip install .

Usage

import renderpy

render_engine = renderpy.Render()
render_engine.setupMesh(MESH_PATH)

camera_model = "OPENCV_FISHEYE"
render_engine.setupCamera(
    height, width,
    fx, fy, cx, cy,
    camera_model,
    params,      # Distortion parameters np.array([k1, k2, k3, k4]) or np.array([k1, k2, p1, p2])
)
near = 0.05
far = 20.0
rgb, depth, vert_indices = render_engine.renderAll(world_to_camera, near, far)

Explanation

render_engine.renderAll render RGB, depth, and corresponding vertex indices of the mesh

rgb: np.ndarray (height, width, 3) uint8
depth: np.ndarray (height, width) float32. Pixels with value 0.0 means no depth
vertex_indices: np.ndarray (height, width) int32. Each pixel represent the three vertex indices in the mesh during rendering. Useful for 2D-3D back projection. Pixels with value -1 means no vertex

Common issues

Missing Dependencies

If some dependencies are missing during build, you can install them by:

# Build tools
apt-get install build-essential cmake git
# OpenGL related
apt-get install libgl1-mesa-dev libglu1-mesa-dev libxrandr-dev libxext-dev
# OpenCV
apt-get install libopencv-dev
# If missing gflags
apt-get install libgflags-dev
apt-get install libboost-all-dev    # for mLib

Version `GLIBCXX_3.4.30' not found

If you encounter missing libstdc++.so.6 error in conda environment, try this:

Add this line to your .bashrc or .zshrc

export LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libstdc++.so.6

fatal error: filesystem: No such file or directory

This is because before gcc/g++ 9, it should be #include <experimental/filesystem> instead (see reference). You can fixed it by replace all the #include <filesystem> with #include <experimental/filesystem> and add link_libraries(stdc++fs) in CMakeLists.txt

Other notes

Currently, the thirdparty/eigen-3.4.0/Eigen has been modified:

MACRO Success is replaced with SuccessfulComputation to avoid conflict with X.h (X11)
See also https://eigen.tuxfamily.org/bz/show_bug.cgi?id=253

liu115 / renderpy