Jiacheng Wang

It seems that there is an additional validation set when performing 5-folds cross-validation.

Firstly, all images are randomly divided into 5 folds. Secondly, in each fold, images are randomly divided into training set and validation set.

For example, let fd1...5 denote the five parts of images. In each experiment fold, four folds are used to form the training set and the rest one is used to...

It seems different with our presented results, BAT(val)

Yes, the pre-trained weight presented here is the fine-tuned version after paper.

Do you process the data by yourself or download the pre-processed data?

Whether the prediction value is NaN?

I have transformed each dataset into the numpy format. In each experiment, the data is splitter into train/validation/test folders, and in each folder, the files are divided into image/label. isi2016...

Please check the shape of your stored label array.

Absolutely yes. Since there remain some details to be improved, we plan to upload the entire project, including the training codes, before the next month.