Help me out!

[OmarAhmadin]

Hello,

Please help me out.
From what I understand; in order to feed my own dataset for semantic segmentation. I have to do the following:

1- I first feed my dataset to the PreprocessingSem3D (using main.cpp) in order to preprocess it.
2- Run train.py.
3- Use predict.py with feeding the best ckpt I have got.

Now, I have multiple questions:

1- Can I use class 0? (I noticed that zero is unlabeled and it is discarded into main.cpp for preprocessing.) I have only two classes. Either powerline or non-powerline points. It should be easier problem. But I did not get any good results.
2- Shall I pass my test data to the preprocessingSem3D before feeding it to prediction.py??
3- After prediction.py I shall feed the output to interplation3D right?

Thanks.

[luoxiaoliaolan]

Hi!
It's my honor to try to answer your questions.

1. I think you can set the non-powerline points to class 0, and the powerline points to class 1.
2. I have preprocessed my dataset(include train data and test data), so I think you should pass your test data to the preprocessingSem3D.
3. Yes, you should feed the output to interplation3D, otherwise you will get the point cloud data after subsampling.