- BlenderProc now can be executed via a python script (experimental, not all modules are supported yet)
- python API extension for the following modules:
- the Entity & Material Manipulation
- the WriterUtility, Hdf5Writer, BopWriterUtility, CocoWriterUtility
- the PostProcessing
- the physic positioning, PoseSampler
- debugging can now be done via the command line, by just typing --debug after a BlenderProc call
- add automatic convex decomposition to speed up physics, object decompositions will be cached
- add a jupyter notebook example for google colab
- add a BasicEmptyInitializer module, which helps creating simple empty objects, these can be used for using a depth of field
- add a depth_of_field option to the CameraInterface (thanks to cuteday)
- add a hide module, so that certain objects can be hidden in a selected range of frames (thanks to Sainan Liu)
- add an option to change the computation type via the config
- blender 2.92 changed depth and distance once again, we use it correctly now and added an explanation
- added a diffuse rendering mode
- added a key frame API
- added a default config, which can be easily changed
- ObjectPoseSampler now can check also against only a limited set of objects
- ObjectReplacer now stops if no objects are left to replace
- SurfaceLighting now reuses the TextureLess material
- preloading in CCTexture now also works perfectly for textures which use an alpha texture
- python packages now are installed by specifying the required packages at the top of each file
- added an example for pasting objects onto random backgrounds
- added a flag to merge BlenderKit objects, while loading, they will be named after the blenderkit file
- reintroduce persistent transformation scaling to improve stability
- improve the clean up at the start of BlenderProc
- fixed a bug in the Front3DLoader, which avoided the correct setting of the custom properties
- fixed a bug with the once_for_all mode for the randomize materials fct.
- add a simple rerun script to show how BlenderProc is used for creating a diverse dataset
- switch to blender 2.92