Didn't work

[Sahil1776]

I have a medical dataset with 3 classes which are slightly different from each other.
I trained ResNet50 for classification and got 74% accuracy on the validation dataset.
I used this ResNet50's weights in train_Food_11_ResNet50.py and tried to get better accuracy.
I got 100% accuracy but combined accuracy didn't exceed 68%. Can you please tell what can I do to improve?
I was hoping to get better results using your approach.

Dataset specifications: X-ray of dog's hearts (Large, Normal and Small categories)