Blenderproc

- segmentations are now done in the same call as any other render call, avoiding the loading of the objects for each pose
- added the kinect azure noise model, allowing for the creation of more realistic depth images
- added a spotlight intersection mode, which places a spotlight in the world without intersecting with the current camera frustum, while it focuses on the middle of the camera frustum, creating hard lighting situations
- add an importer for the MatterPort3d dataset
- the key of custom properties now can no longer have the same name as any attribute of the `blender_obj`. Custom properties do not need to start with `cp_` anymore
- add a new paper designated for the journal of sensors
- we now use pylint to clean up all code smells and add documentation to all functions
- the bop loader now also supports the hope dataset
- separate the init and clean up functions
- it can now be specified which GPU to use for rendering, if multiple are available
- rewrite the ambient CG download script, allowing resuming the download
- turn of the cycle denoiser causing issues in certain settings
- add a random walk feature, which allows simulating a camera shaking or POI drift
- the duplicate and delete fcts now support duplicating all children as well, we added a `get_children` fct as well
- renaming `get_rotation` to `get_rotation_euler` and adding functions for setting the rotation matrix
- fix a bug in the move_origin_to_bottom_mean fct.

- allow writing poses for robot links in the BopWriter
- loading .obj files now uses the faster importer
- fix a pip install bug
- fix a bug in the LinkUtility of the URDF loader

- add urdf support:
- this enables the simulation of robotic arms with forward and inverse kinematics
- add an example to show of how this works
- add a new face slicer option, making it easier to slice the top of a table away for placing objects on them
- add a new example for sampling objects in 3D Front scenes
- switch to vhacd version 4.X allowing for faster decompositions than before, speeding up the simulation
- add an option to link blend files instead of loading them (faster, but objects can not be changed after linking)
- add a new replica loader, loading semantic objects instead of just one mesh for the whole scene
- add basic support for apple silicion (M1, M2), this might still mean that some packages have to be installed on their own afterwards
- add return random material to the load haven material function, the function returns also now the full loaded list of materials
- add new function to create a material based soley on a texture (Path or bpy.types.Image)
- upgrade to blender 3.2.1
- reduce the memory demand of the semantic segmentation by fixing a small bug
- fix a bug where the coco annotation writer wouldn't work if the frame_start is not zero
- fix a bug where writing alpha images in the bop writer did not work
- fix a bug where pip would reinstall packages all the time
- fix a bug where on windows the pip path would be wrong

- upgrade to blender 3.1.0
- add support for Apple Silicon and GPU on Mac OS 12.3
- add new stereo image projector, making it easier to simulate SGM depth
- add quickstart command `blenderproc quickstart`
- add gif writer
- function to scale the UV texture coordinates of a `MeshObject`
- `object.sample_poses()` now also supports a `mode_on_failure`, which allows to control where the object is placed if the sampling fails marcelhohn
- improve error message for python 2.X
- bug fixes:
- `set_origin()` doesn't change the cursor position anymore marcelhohn
- `vis hdf5 --save` now also supports stereo images
- improve error message when rendering with no camera poses
- fixed a bug where the windows path was incorrect for blender marcelhohn
- fixed a bug when the category name is an int during coco vis
- `load_haven_mat()` now loads more materials, not all materials have been used before Victorlouisdg

- switch blender version to 3.0.0 instead of 2.93.0
- we now rely on Cycles X, making the rendering much faster than before
- this also depreactes the usage of `bproc.renderer.set_samples()`, this is now replaced with `bproc.renderer.set_noise_threshold()`. This fcts allows to set the desired noise ration on a pixel basis, giving a much higher control to ensure a certain consistent noise level in the whole image.
- it is still possible to limit the amount of samples per pixel with a new function named: `bproc.renderer.set_max_amount_of_samples()`
- as the whole image is now rendered at once we removed the auto-tile addon
- the BLENDER denoiser is no longer available, we recommend using the INTEL denoiser.
- remove the argument `keep_using_base_color` from `bproc.lighting.light_surface()` and `Material.make_emissive()`, now either a emission color or the base color is used
- changes for the bop integration
- install bop_toolkit automatically
- add BlenderProc2 python scripts for BOP challenge
- allow to pass list of objects to `write_bop` for which to save pose annotations
- divide `load_bop()` into `load_bop_objs()`, `load_bop_scene()` and `load_bop_intrinsics()` to decouple the functionalit
- fix a bug after uninstalling pip packages they were not truly removed

- add new lens distortion module, adding the possibility to simulate `k1, k2, k3, p1` and `p2` parameters.
- improve usability of distance rendering, make it the default in all examples
- distance and depth rendering can now be done antialiased and non-antialiased
- add a nocs renderer: [paper](https://geometry.stanford.edu/projects/NOCS_CVPR2019/pub/NOCS_CVPR2019.pdf)
- upgrade the shell sampler and rework the usage and documentation
- move `run.py` to `cli.py` and support the same CLI arguments as for `blenderproc ...`
- make caching of pip installs the default (faster start-up) and add option to reset them if needed
- refactor the vishdf5 script using the same content now for saveAsImg
- clean up the scripts folder and move it completely into blenderproc
- include all packages (like vhacd) in the pip install package
- fix bug with light internal blender_obj after an undo operation
- add OPTIX as denoiser option
- add possibility to read `jpeg`
- fix `use_ray_trace` in `bproc.sampler.upper_region`
- added an exception if not the correct python environment is used