Identify potential animals and crop to them before applying classifier

[gsganden]

Rather than providing entire images to a classifier, it might work better to identify areas of interest as a first step and apply the classifier within those areas.

[AnshuTrivedi]

@gsganden How it will work if area of interest is more then one in image like cat and dog classification where we look for different features at different parts of image?Specially selecting more then one area of interest in a single image.

[gsganden]

• It is quite rare to have multiple types of animals in the same image, so we can probably get away with not worrying about the case, at least on a first pass.
• It is fairly common to have multiple instances of certain types of animals (e.g. raccoons) in one image. Some kind of averaging or similar over multiple detected areas should work pretty well, I think.

Does that help? What do you think?