Analysis: training samples v/s performance

[sedflix]

The goal here is to see how many training samples are required to get good enough performance.

On top of my head, these would be the required approach:

• Take 5 images: 15.jpg to 20.jpg images as a test set
• Train the model, with 5 images(1-5) and 10 image(1-10).
• Report the result on the test set!

Let me know if this is fine?