search

Hengwei-Zhao96 / T-HOneCls

Class prior-free PU Learning for hyperspectral imagery (ICCV 2023)
22 stars 0 forks source link

About the dataloader #2

Closed BoYangXDU closed 1 month ago

BoYangXDU commented 3 months ago

Hengwei,

Thank you for your valuable contributions to PU learning for hyperspectral image analysis, particularly with absNegative, HOneCls, and T-HOneCls.

I'm trying to follow your work, but I've noticed some differences in the data loading files for the three datasets. Specifically, for dataset_HongHu, there are the following lines: https://github.com/Hengwei-Zhao96/T-HOneCls/blob/7cf2de6668a4a7ca5b1812f3ea42865355940152/HOC/datasets/dataset_HongHu.py#L32-L35

However, I couldn't find similar code for dataset_HanChuan and dataset_LongKou. Could you please explain the differences between these configurations?

Thank you very much.

Best regards, Bo

Hengwei-Zhao96 commented 1 month ago

Howdy,

Apologies for the delayed response!

The difference arises from how the labels in these hyperspectral datasets are structured. In the HongHu dataset, both the training and testing labels come from the same file, whereas in the other two datasets, they are separated. As a result, we need to configure the process differently.

Best,

Hengwei Zhao

BoYangXDU commented 1 month ago

With great thanks for your help. It is really helpful. I will close this issue. Best.

Yang