Skip to content

README.md

Latest commit

 

History

History
82 lines (62 loc) · 2.51 KB

README.md

File metadata and controls

82 lines (62 loc) · 2.51 KB

Buildee: A 3D Simulation Framework for Scene Exploration and Reconstruction with Understanding

Paper • Project page

This repository hosts Buildee, a 3D simulation framework built as a Python module on top of Blender. With Buildee, you can explore a 3D scene and generate realistic RGB, depth, and semantic segmentation maps. Buildee is also equipped to track 2D / 3D points and perform occlusion checking.

buildee_features.mp4

Installation

Clone the repository with submodules:

git clone --recursive https://github.com/clementinboittiaux/buildee.git

Setup the conda environment:

conda create -n buildee python=3.11
conda activate buildee

Install the numpy version required by blender:

pip install numpy==1.24.3

With the conda environment activated, build blender as a python module:

cd buildee/blender
./build_files/build_environment/install_linux_packages.py  # update dependencies
./build_files/utils/make_update.py --use-linux-libraries  # update dependencies
make bpy  # build bpy module
python3 ./build_files/utils/make_bpy_wheel.py ../build_linux_bpy/bin/  # create wheel

Then, install blender's bpy python module:

pip3 install ../build_linux_bpy/bin/bpy-4.4.0a0-cp311-cp311-manylinux_2_35_x86_64.whl

Install your compatible PyTorch version. Then install other requirements:

cd ..
pip install -r requirements.txt

Finally, install Buildee as a package:

pip install -e .

Troubleshooting

When installing on a headless server, remember to disable X11 forwarding.

Usage

With Buildee, you can start your simulation in just a few lines of code. The following example shows how to:

  1. Load a Blender scene.
  2. Render RGB image, depth map, and segmentation map.
  3. Compute the current point cloud of the scene (dynamic points positions may change over time).
  4. Step forward the simulation to the next frame.
from buildee import Simulator

sim = Simulator(
    blend_file='file.blend',
    points_density=10.0,
    verbose=True
)

rgb, depth, labels = sim.render()

pcl, pcl_labels, pcl_visibility_mask = sim.compute_point_cloud(imshow=True)

sim.step_frame()