BlenderProc2 for 3D-Front

Support BlenderProc2 with multi-GPU batch rendering and 3D visualization for the 3D-Front dataset.

If you feel struggled in rendering 3D-Front or visualizing its 3D assets for your tasks. Here is a solution based on BlenderProc2 and VTK.

2D rendering

RGB	Depth	Semantics	Instances

3D visualization

Pointcloud	Layout & Camera Poses	CAD models + Oriented Bounding Boxes

Note:

X server is required for 3D visualization.
OpenGL camera setting is used.

Please check the orginal REAME if you want to know all the functions in BlenderProc2.

Install

Use conda to deploy the environment by

cd BlenderProc-3DFront
conda env create -f environment.yml
conda activate blenderproc
pip install -e .

Apply for the 3D-Front dataset. Download all the data and link them to the local directory as follows:

examples/datasets/front_3d_with_improved_mat/3D-FRONT
examples/datasets/front_3d_with_improved_mat/3D-FRONT-texture
examples/datasets/front_3d_with_improved_mat/3D-FUTURE-model

Download textures data from link by

blenderproc/scripts/download_cc_textures.py

Render

Single scene rendering

Here we take the scene ID 6a0e73bc-d0c4-4a38-bfb6-e083ce05ebe9.json as an example. We can do multi-view renderings by running the script as follows:
```
blenderproc run \
examples/datasets/front_3d_with_improved_mat/render_dataset_improved_mat.py \
examples/datasets/front_3d_with_improved_mat/3D-FRONT \
examples/datasets/front_3d_with_improved_mat/3D-FUTURE-model \
examples/datasets/front_3d_with_improved_mat/3D-FRONT-texture \
6a0e73bc-d0c4-4a38-bfb6-e083ce05ebe9.json \
resources/cctextures/ \
examples/datasets/front_3d_with_improved_mat/renderings
```
- The rendering results will be saved in examples/datasets/front_3d_with_improved_mat/renderings.
- You can also customize your camera FOV, rendering resolution and number of views, etc. in render_dataset_improved_mat.py. Just dive deeper and play with it.
- You can also output normal maps or depth maps by uncomment bproc.renderer.enable_normals_output() and bproc.renderer.enable_depth_output(activate_antialiasing=False).
- I customized the camera pose sampling function to support multi-view renderings. The sampling strategy ensures every camera pose has a reasonable object surface coverage.
- The number of camera poses in a room is proportional to the ratio of its floor area to the whole apartment. Default camera pose number is 100 per apartment.
Batch scene rendering (support multi-GPU mode for parallel rendering)

This script is built on the above function to support batch rendering. Run the following script to render all scenes in a loop.
```
python examples/datasets/front_3d_with_improved_mat/multi_render.py \
examples/datasets/front_3d_with_improved_mat/render_dataset_improved_mat.py \
examples/datasets/front_3d_with_improved_mat/3D-FRONT \
examples/datasets/front_3d_with_improved_mat/3D-FUTURE-model \
examples/datasets/front_3d_with_improved_mat/3D-FRONT-texture \
resources/cctextures/ \
examples/datasets/front_3d_with_improved_mat/renderings \
--n_processes 1
```
- The rendering results will be saved in examples/datasets/front_3d_with_improved_mat/renderings.
- You can still customize your rendering pipeline by modifying render_dataset_improved_mat.py.
- If you have multiple GPUs and want to render in parallel, change n_processes to your GPU number.

Visualization

After rendering a scene, please run the following code for 2D and 3D visualizations

python visualization/front3d/vis_front3d.py --json_file 6a0e73bc-d0c4-4a38-bfb6-e083ce05ebe9.json

Citation

Some parts of this code base is developed based on the following works. If you find our work helpful, please consider citing

@inproceedings{nie2023learning,
  title={Learning 3d scene priors with 2d supervision},
  author={Nie, Yinyu and Dai, Angela and Han, Xiaoguang and Nie{\ss}ner, Matthias},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={792--802},
  year={2023}
}

@article{Denninger2023, 
    doi = {10.21105/joss.04901},
    url = {https://doi.org/10.21105/joss.04901},
    year = {2023},
    publisher = {The Open Journal}, 
    volume = {8},
    number = {82},
    pages = {4901}, 
    author = {Maximilian Denninger and Dominik Winkelbauer and Martin Sundermeyer and Wout Boerdijk and Markus Knauer and Klaus H. Strobl and Matthias Humt and Rudolph Triebel},
    title = {BlenderProc2: A Procedural Pipeline for Photorealistic Rendering}, 
    journal = {Journal of Open Source Software}
} 

@InProceedings{Paschalidou2021NEURIPS,
  author = {Despoina Paschalidou and Amlan Kar and Maria Shugrina and Karsten Kreis and Andreas Geiger
  and Sanja Fidler},
  title = {ATISS: Autoregressive Transformers for Indoor Scene Synthesis},
  booktitle = {Advances in Neural Information Processing Systems (NeurIPS)},
  year = {2021}
}

Yannnnnnnnnnnn / BlenderProc-3DFront