Comparing face shape observations

[LucasOa]

I have been asked to suggest test cases for the face outline step in a photo form. I want to distinguish a different crop from an actual change in the visible facial outline. I would begin with a single authorized image and record the dimensions before making any copies. Each copy would retain the same facial pixels while adding or removing empty space around them. I am considering face shape detector as a reference while drafting the questions. The plan is still on paper, with no measured outcomes or confirmed behavior behind it. For each copy, I could note its border dimensions, category and any confidence information provided. Would you compare only the final categories, or also record where the outline was detected? My priority is a clear account of changed outputs, with room for genuinely uncertain classifications.