How are the segmentation masks obtained?

Question

How are the segmentation masks obtained?

Opened this issue 3 years ago · 4 comments

Hello,
thank you for the great work!!
I want to try Neural-PIL on my own dataset, could you please let me know how were the segmentation masks generated?

Thank you for your time.

Answer 1 · 2022-03-13T19:00:50.000Z

I am not associated with the authors, but I have trained the model recently on a custom dataset. You can try using U2Net which can be found at https://github.com/xuebinqin/U-2-Net. It has worked quite well for me.

Have you figured out how to get the decompositions by running the code? I was only able to train.

Answer 2 · 2022-03-14T01:03:53.000Z

Hi @ArenaGrenade, Thank you for the suggestion, U2Net looks great!
-> "Have you figured out how to get the decompositions by running the code? I was only able to train."
I just finished training the network on example data (Gnome), and am about to check how to get the decompositions. I will come back to the discussion if I figured it out.

Answer 3 · 2022-04-04T12:27:10.000Z

Any update on decompositions? I'm also working on the same problem. Is there code for decomposition of the original NeRD model?

Answer 4 · 2023-02-18T00:59:53.000Z

I am not associated with the authors, but I have trained the model recently on a custom dataset. You can try using U2Net which can be found at https://github.com/xuebinqin/U-2-Net. It has worked quite well for me.

Have you figured out how to get the decompositions by running the code? I was only able to train.

It seems that the outputs include the decompositions of the pictures in training datasets and a few additional test images' decompositions decided automatically. But I don't know how to generate a new picture's decompositions with the trained model. Do you find any solution in the end?