the difference between the number of classes

[ayoubOUSSM2021]

Hi ; I tried to apply the model on my dataset, I prepared this data as Semantic3D dataset but I have 6 classes instead of 8 classes and I set the number of classes = 6 in helper_tool.py file but when I run the training the metrics = NaN
Can you help me to solve this problem and thanks in advance

[loki-veera]

Is your problem solved, because I am also facing same issue with semantic KITTI dataset. If yes, it would be helpful to share the solution.

[ayoubOUSSM2021]

Hi , I have not yet solved the problem and you ?

[loki-veera]

I kind of solved it but the issue is with new classes the network doesn't optimize. I changed the semantic KITTI classes from 19 to 17 and also changed the https://github.com/QingyongHu/RandLA-Net/blob/6b5445f5f279d33d2335e85ed39ca8b68cb1c57e/helper_tool.py#L256 by removing the weights of classes which are not needed.

[ayoubOUSSM2021]

Hi , I've done this before but still the problem of NaN , please see screenshot (I have 6 classes )

[loki-veera]

I think the problem is with division if your denominator is zero then leads to infinity which leads to NaN. I would try adding a small constant of 1e-6 to denominator (like float(expression+1e-6)) and run. Please let me know the status if you run like this.

[ayoubOUSSM2021]

I tested this but unfortunately still the same problem, please if you have other ways to solve the problem and thank you in advance I think the problem is in this part:

[ayoubOUSSM2021]

Hi , Is your problem solved ? , if yes, it would be helpful to share the solution. thank you in advance

[loki-veera]

Hi Unfortuntely no, I didn't solve it.