second try of finetuning, trained for 50 epochs without data augmentation (pretrained weight for the entire model)