Does anyone know how to adjust the class_weight to a list for training on a dataset with extremely imbalanced label pixels?

[jason-60904]

I am training a human segmentation model, and the pixel distribution in the dataset is extremely imbalanced.

However, due to the complexity of this project, I find it difficult to modify it to accept class_weight as a list.

Does anyone have any suggestions?

[dou3516]

same question