You have four possibilities:
- PacWGAN-GP2 (used in original DeepIE3D)
- WGAN-GP
- PacDCGAN2
- DCGAN
Follow this guide:
- Get some data from ShapeNet - download in voxel format
- Downscale the voxel models to 64 X 64 X 64 or make changes to model.py - script to downscale is included: downscale.py
- Place the downscaled models in a folder named "data"
- You're now ready to train! To configure hyperparameters, please refer to argparser.py
- If you don't wanna do the above, some training data has already been included; running main.py will work right away.
The trained model is usable together with the DeepIE3D_Backend