search

Daniel-Liu-c0deb0t / 3D-Neural-Network-Adversarial-Attacks

Research on adversarial attacks and defenses for deep neural network 3D point cloud classifiers like PointNet and PointNet++.
23 stars 8 forks source link

Do I need to train for Object part Segmentation in pointnet #5

Open binbbaz opened 3 years ago

binbbaz commented 3 years ago

I have been training pointnet for the past three days as a requirement to run this code. I just finished running train.py in the home directory of pointnet. Do I need to train for object part segmentation as well before moving on to pointnet2?

Daniel-Liu-c0deb0t commented 3 years ago

No, only classification is attacked with this code. Did you make the modifications to PointNet code as suggested in the readme?

binbbaz commented 3 years ago

No, I have not. I am still facing some challenges in training pointnet2. I will make the suggested modifications after training pointnet2

Daniel-Liu-c0deb0t commented 3 years ago

Might be important to note that it is probably difficult to get this code to work out-of-the-box due to weird setups I had before. I mentioned this in the readme, but a continuation is available here, which has cleaner code and more detailed setup instructions (PointNet/PointNet++ would be trained the same way). I didn't reimplement all the attack algorithms in the new code, only a couple important ones that I wanted to evaluate.

binbbaz commented 3 years ago

Thank you. I will be sure to check it out and ask for help when necessary.

Daniel-Liu-c0deb0t commented 3 years ago

I'm happy to help if you have any other questions.