Blenderproc

Latest version: v2.8.0

Safety actively analyzes 715032 Python packages for vulnerabilities to keep your Python projects secure.

Scan your dependencies

Page 4 of 6

1.9.0

- all `Loader` now support setting the `add_material_properties` of the newly loaded objects
- each new `.hdf5` file, will contain the git commit hash of BlenderProc
- introduce `LightUtility` -> this fulfills our long time goal giving BlenderProc a python API
- changed the default stereo mode from `OFFAXIS` to `PARALLEL`
- adapt to the changes made to 3D Front, as 3D Front does not have a version number, we support the newest version and a version from Summer 2020, if you have errors please open an issue
- `getter.Material` can now return all materials of a list of objects, selected via the `getter.Entity`
- add a `cf_add_*` fct to the MaterialManipulator, which works similar to the `cf_set_*` fct.
- add `check_empty` to `getter.Entity` and to `getter.Material` to throw an error if the returned list is empty
- added a `cf_add_uv_mapping/forced_recalc_of_uv_maps` option to force to recalculate the uv map of materials, which already have a uv map
- added option that randomized materials in `cf_randomize_materials` are added to objects without any materials
- change that `cf_randomize_materials` inside the `EntityManipulator` now deals with lists and not single elements as before
- add auto download for the imageio library, if that fails an exception is thrown with instructions on how to do this yourself
- `CameraInterface` now supports setting poses for a certain frame -> also needed for the python API
- changed the `depth_scale` default value in the BopWriter from `0.1` to `1.0`, added an option to change via the config
- changed the `ignore_dist_thres` default value in the BopWriter from `5.0` to `100.0`
- improved the ShapeNet example, by adding a ShapeNetWriter, which saves which object was used in the `.hdf5` container
- fixes a bug with the separation of ceilings and floor objects in SceneNet scenes
- fix a bug, where the `add_alpha_channel_to_texture` fct. couldn't deal with empty material sockets
- fix the `pyhsics_positioning` example by using the `collision_shape="MESH"`
- fix a bug, where empty material_sockets would have been returned in the `getter.Material`
- fix `vis_coco_annotation` script, didn't work anymore after the segmentation map rewrite

1.8.2

- added stereo mode to SegMapRenderer
- switched to using imageio for reading images, as the blender image loading API does not support .exr files written in stereo mode
- added option to write world-to-cam transformations to the BopWriter
- CocoAnnotationWriter does now in the polygon format not write empty segmentation lists anymore and uses iscrowd:0
- type hints are now added to the generated documentation
- extracted reusable rendering code from the renderer modules into utility classes

1.8.1

- fixed a bug in the WriterInterface
- fixed cc_texture downloader script in the case of weird urls
- fixed creating the ceiling when there are rounding errors in the vertices coordinates

1.8.0

- massively improve the documentation, by adding using github pages for our documentation: https://dlr-rm.github.io/BlenderProc/index.html
- add a RandomRoomConstructor:
- This module can generate random floor plans of single rooms, with arbitrary extrusions.
- It can also randomly place loaded objects without collision inside of the room
- By using CCTexture it is also possible to assign random materials to the floor, wall and ceiling
- remove the SceneNetLighting module and replace it with the SurfaceLighting module, to make it more general
- add support for the haven websites and add an example and corresponding loaders:
- The haven environment website: https://hdrihaven.com/
- The haven model website: https://3dmodelhaven.com/
- The haven texture website: https://texturehaven.com/
- add option to the CameraSampler to ensure that a certain object is always in the camera view
- improve the FloorExtractor to extract the floor and ceiling in SceneNet and other scenarios
- improve the skin tones for AMASSLoader
- add a proper scaling to the ikea dataset, by converting all files into SI units
- fix the ikea dataset downloading, by removing and splitting broken pieces, we advise to redownload the dataset

1.7.0

- switch to blender 2.91
- added an example of how to set object poses via a transformation matrix and set camera extrinsics / intrinsics via a transformation matrix and a K matrix
- added camera utility class which makes it easier to set and retrieve intrinsics via any K matrix
- added loader for Pix3D dataset
- added loader for AMASS dataset
- added loader for the ikea dataset
- added community driven support for Windows
- added motion blur and rolling shutter support
- fixed collision checks between meshes, so the ObjectPoseSampler is not generating colliding object poses anymore
- fixed cleanup of temporary directories in the case of an error
- set pixel origin to the center of a pixel, slightly affecting intrinsics and depth outputs
- fixed coco annotations if background is not visible
- added support for transparent background
- fixed wrong image size in coco annotations
- fixed blender proc if non-english language is configured
- when loading ply files a default material is now added
- fixed setting matrix_world via the entity manipulator
- fixed the scenenet examples (corrects physics positioning and camera sampling)

1.6.1

- fixed bbox computation in the Coco Annotations
- fixed visualization of Coco Annotations
- add set properties fct. to BlenderLoader
- adds a script to download all materials from BlenderKit (not used yet)

Page 4 of 6

© 2025 Safety CLI Cybersecurity Inc. All Rights Reserved.