This repository contains code to render the ShapeNet and ABC datasets from various poses, lighting and material properties.
Note: there is overlap in the code for ShapeNet and ABC but there are subtle differences that would make a unified codebase more complex without significant gain.
Download Blender 2.79b for your operating system from the official Blender release.
For MacOS the default installation path is
/Applications/Blender/blender.app/Contents/MacOS/blender
,
and for Linux it involves extracting a blender-2.79b-linux-glibc219-x86_64.tar
which makes the path
<path to extracted tarball>/blender-2.79b-linux-glibc219-x86_64/blender
.
Depending on your OS, add the appropriate path in the data_generation_parameters.json
file under the key "blender_path"
.
The file data_generation_paramters.json
is used for both ABC and ShapeNet.
- Download the datasets from their respective locations
- Further instructions are included in each dataset directory
- Postprocess the rendered data to turn the
.exr
depth outputs into images andmin
andmax
values for the depth on the visible object surface usingutils/convert_exr.py
. This provides absolute depth information.