i’m currently working on image classification using Resnet50 on the Amazon Berkley Objects Dataset, i’ve been facing the issue of a low F1 score, i made sure the classes are equal in both the training and testing sample (a total of around 50k images), despite that it wont go past 10% (ik the image is showing the 7th epoche but i wouldnt budge till the end of the runtime), any suggestion?

the preprocessing steps are mainly filtering, data augmentation, rescaling, basic data preparation, the data is split 80 20.