06. Region Proposals
05 Region Proposals V1 RENDER V2
Suggested region proposals for an image of a cat and dog.
Reflect
QUESTION:
Consider the above image, how do you think you would select the best proposed regions; what criteria do good regions have?
ANSWER:
The regions we want to analyze are those with complete objects in them. We want to get rid of regions that contain image background or only a portion of an object. So, two common approaches are suggested: 1. identify similar regions using feature extraction or a clustering algorithm like k-means, as you've already seen; these methods should identify any areas of interest. 2. Add another layer to our model that performs a binary classification on these regions and labels them: object or not-object; this gives us the ability to discard any non-object regions!