im using tensorflow v2 from the tutorial api and im training a pre trained model resnet50 however, with 60k steps or epochs, i exported the model and test it the result where inaccurate as it detects the image but the class or label is wrong, some classes are correct but most often not, for context my dataset is Philippine peso bills, and errors such as classifying 500 peso bill as 100 peso bill. Any suggestions i could do?
Ive tried changing the learning rate to make it decrease slower. And i've added many images in dataset. I also have checked the label and it is correct, even the labelmap is correct