Code Repository to check robustness of 3D Deep Learning (Volumetric and PointNet) to occlusion attacks. This repository reproduces the results reported in the CVPR2019 paper:
'Robustness of 3D Deep Learning in an Adversarial Setting'
This repository also contains all of the code and data for completing the following tasks:
- Training of PointNet and VoxNet architectures on ModelNet and KITTI
- Evaluating trained architectures with random and saliency occlusion
- Plotting/Displaying 3D files from the dataset + their critical sets
- Reproducing Plots from the paper.
Things that can be requested via email:
- Code to generate KITTI object dataset used by this paper
- Original driving sequences used to source the KITTI dataset