Cow teat dataset contained the following class distribution before Stable Diffusion which shows a very high data imbalance between classes. That's why Data Augmentation was important to generate images for underrepresented classes such as 1, 4, and 3.
Here is the data distribution of the classes:
Score_1: 450, Score_2: 491, Score_3: 187, Score_4: 21
Tested a total of 6 models on both the balanced and unbalanced datasets. The paper can be found here
Please reach out to me for any questions or collaboration.