Training on custom datasets in CCT.

[TinaWut]

First of all, thank you very much for sharing your paper and code. I have also read the previous answers regarding issues with training on custom datasets, but I still haven't resolved the problems I encountered during my own dataset training.

My dataset has two classes (0 for background, 1 for another category). I modified num_classes = 2 in voc.py and also adjusted the color mapping in palette.py with def get_voc_palette(num_classes) to represent two classes (black and white). I am training using a supervised approach, but I am encountering a situation where the intersection over union (IoU) for one of the classes is 0. Additionally, when loading the same model weights in inference.py, the output results vary with each run. I have no idea how to resolve these issues. Any guidance would be greatly appreciated

[TinaWut]

[TinaWut]

The code issue has been resolved

[yassouali]

Hi @TinaWut, happy to hear that the problem was resolved, I'll close the issue for now

[79zzz]

您好，请问你是如何自定义数据集的