This is a Python library for constructing a dataset for GQN and CGQN. This renderer has two backend implementations available: OpenGL backend and CUDA backend.
- Python 3
- Pillow
pip3 install Pillow
- tqdm
pip3 install tqdm
- Ubuntu
- tested on Ubuntu 16.04 / 17.10
- GPU
Also, you need Chainer, PyTorch or Keras to download MNIST images.
pip3 install -r requirements.txt
cd opengl
python3 shepard_metzler.py --num-cubes 7 --num-colors 10 --output-directory shepard_metzler_7_part --total-scenes 2000000
cd opengl
python3 rooms_ring_camera.py --output-directory rooms_ring_camera_no_object_rotation --anti-aliasing
cd opengl
python3 rooms_free_camera.py --output-directory rooms_free_camera_no_object_rotation --anti-aliasing
cd opengl
python3 mnist_dice_ring_camera.py --output-directory mnist_dice_ring_camera_no_object_rotation --anti-aliasing
- Python 3
- OpenCV
pip3 install opencv-python
- Pillow
pip3 install Pillow
- tqdm
pip3 install tqdm
- pybind11
pip3 install pybind11 --user
- Ubuntu
- tested on Ubuntu 16.04 / 17.10
- CUDA
- tested on CUDA 9.1
- NVIDIA GPU
- tested on GTX 1070 / 1080
- C++14 (gcc-6)
Also, you need Chainer, PyTorch or Keras to download MNIST images.
cd cuda
make -j4
cd cuda
python3 shepard_metzler.py --num-cubes 7 --num-colors 10 --output-directory shepard_metzler_7_part --total-scenes 2000000
cd cuda
python3 rooms_ring_camera.py --output-directory rooms_ring_camera_no_object_rotation --anti-aliasing
cd cuda
python3 rooms_free_camera.py --output-directory rooms_free_camera_no_object_rotation --anti-aliasing
cd cuda
python3 mnist_dice_ring_camera.py --output-directory mnist_dice_ring_camera_no_object_rotation --anti-aliasing