Visualizing Global Explanations of Point Cloud DNNs

This work is based on existing studies: the generative model is based on this repo, the classification model is based on this repo. Please build the environment according to the corresponding requirements.

Usage:

1. Train PointNet to have a classification model to be explained first, follow the step of this repo.
2. Train our 3 different kinds of generative model via PointCloudAE.py, AED.py and NAED.py
3. Run batch AM process on ModelNet40 for corresponding generative models via vanilla_VAE_AM_batch.py, AED_AM_batch.py and NAED_AM_batch.py
4. You can evaluate AM examples via different metrics: Chamfer, EMD, FID, modified Inception Score and our Point Cloud Activation Maximization Score via Chanfer_eval.py, EMD_eval.py, FID_eval.py, m_IS_eval.py and AM_eval.py