I have 4 classes and each class contains roughly between 20-30 images.
My first thought is to get more data but what can be done if I can’t get more data?
No, this doesn’t look like overfitting. It looks more like something else went wrong. Your error rate shouldn’t be worse than random chance.
Have a look at your data and make sure it’s getting loaded and labeled correctly. Double check your setup code. And try adjusting your learning rate.
What are your thoughts on working with the small dataset? What things I need to look out for when my dataset is very small?
It probably won’t generalize very well. But without knowing any more about your dataset, my guess is that it should be working better than the results from your first post so something else is likely going wrong.
If you provide more details (like a link to a notebook) people will probably be able to provide more specific guidance.