Supervised Learning with predefined ROI


I want to predefine the ROI that contains defects in my data before training them. My model extracts irrelevant features that I don’t want, so with predefining the ROI I want to make these defects more obvious for my model. In is there any tool/method exists for supervising the ROI before training the model?

Why don’t you crop your Image and just keep the part with the defect? Also I think you should mention what type of problem do you have, is it an Object-detection problem or Image classification problem.

It’s a classification problem, and the defect does not have certain geometrical shape. I have already cropped the defects but as the defect does not have a certain geometrical shape, the cropped region includes some of the unwanted regions.

Let’s say you succeed in providing ROI. How are you gonna do that for test images?