Blender script for rendering ShapeNet models

This repo contains the implementation of Blender Python script for rendering ShapeNet obj files

example:

rendered rgb images with random background images

rendered depth images:

pose:

4.374694554125504453e-01 -4.180723275263726890e-01 7.961381881998743637e-01 1.691706400439716873e+00
8.974623655743290129e-01 1.474515273040246655e-01 -4.157154855070874788e-01 -8.246252502494784142e-01
5.640734111112344351e-02 8.963668844113694689e-01 4.397097116636921044e-01 9.232025044714089468e-01
0.000000000000000000e+00 0.000000000000000000e+00 0.000000000000000000e+00 1.000000000000000000e+00

this matrix contains rotation matrix and translation vector, transforms coordinates from object coordinates to camera coordinates(actually is original camera pose to rotated camera pose since Blender computes everything in world coordinates system). I've verified the rotation matrix using CPP Eigen library, Python Tansforms3d library, CPP Opencv library, but if im wrong, plz let me know.

useage for Python code

0. requirement

Python 3.5

Blender 2.79

ShapeNet Dataset(if not, you need to modify the code yourself)

background images(recommend PASCAL VOC)

1. configuration file

you need to change the settings.py file before you run the python file!!!

all the configurations needed are in '''settings.py''', you need to set your own:

• ShapeNet dataset path: g_shapenet_path
• Blender executable path: g_blender_excutable_path

and other configurations you might want to change:

• background image folder path: g_background_image_path
• synthetic rgb image folder path: g_syn_rgb_folder
• synthetic depth image folder path: g_syn_depth_folder
• pose folder: g_syn_pose_foloder

for more configuration informations, plz refer to Blender API Document

2. viewpoint file

All txt files in the folder view_points are viewpoints file, you can generate new viewpoints file using render for cnn

• you can change the the camera clip distance by modifying varibale distance_min and distance_max in RenderForCNN/blob/master/render_pipeline/kde/sample_viewpoints.m,

• and view point number generated per categlory by modifying num_samples in RenderForCNN/blob/master/render_pipeline/kde/run_sampling.m

3. start rendering

python run_render.py

this command will render the rgb, depth image and generate pose file.