BlenderFunc

Online Documentation

This repo was inspired by BlenderProc: https://github.com/DLR-RM/BlenderProc.git

The differences between this project and BlenderProc:

• Instead of writing config in BlenderProc, you can directly write python script to control the data generation process, which gives you more flexibility

• Simple but powerful functions, e.g., set_camera(), set_projector(), add_ply(), physics_simulation(), render_color()

How to run

cd BlenderFunc

## install Blender 2.92
wget https://download.blender.org/release/Blender2.92/blender-2.92.0-linux64.tar.xz
# or use Tsinghua source
# wget https://mirrors.tuna.tsinghua.edu.cn/blender/blender-release/Blender2.92/blender-2.92.0-linux64.tar.xz
tar xvf blender-2.92.0-linux64.tar.xz
ln -s blender-2.92.0-linux64/blender .

## run helloworld.py
./blender --background --python examples/helloworld.py

Use docker to generate xyz_log

## build docker
./docker/build.sh

## run docker
# help
./docker/run.sh --help

# generate data to specified directory (WARNING: please use absolute path)
./docker/run.sh --output_dir=$(pwd)/xyz_log

# for convenience, you can copy this script to /usr/local/bin
sudo cp ./docker/run.sh /usr/local/bin/synthesize_xyz_log
synthesize_xyz_log --output_dir=$(pwd)/xyz_log

Todo

features

• shadow map
• add_simple_texture
• support BW output
• material location, rotation, scale, main color
• save_blend_file for each renderer
• class segmap
• instance segmap
• normal
• pose output
• hdr downloader
• hdr texture
• camera distortion
• projector distortion
• docker image
• ABC dataset download
• ABC dataset process
• calibration pipeline
• support stl model
• upgrade collision checking
• pose sampler return same format
• xyz-log format
• coco format
• documentation

bugs

• physics simulation origin wrong
• fail to enable GPU
• set object.name to an existed name