Training issue about jcs HOT 4 OPEN

Thanks a lot for your enquiry. train_single.py is to train the segmentation branch. During the joint training stage, you could load the trained weights of segmentation branch (pretrained on segmentation maps) and the classification branch (pretrained on image-level maps) separately, then conducting the joint training (finetuning).

from jcs.

Thank you for your response. I need assistance with two issues:

1. How can I train a classification branch using techniques like image mixing and pixel-level supervision for activation mapping, as mentioned in the paper?
2. Is the code for this specific part, especially for training a classification branch on a custom dataset, available in this GitHub repository?

from jcs.

We have no plans to release the entire code of training the classification branch currently.
However, you can reimplement the training process as all details (process, hyper-params) have been explained clearly in our paper.
Thank you for your understanding.

from jcs.

Thanks for your clarification. I appreciate your great work. Have a nice day!

from jcs.