Best way to perform categorization of multiple objects in one single picture

Part 1 (2019) Advanced (Part 1 v3)
1

Hi,
I have the following scenario. I succesfully trained an image classifier to identify black bears and grizzlies.
I do not have a training set of bears with bounding boxes.
Imagine that I have now a picture containing 2 grizzlies and 1 black bear. Is there a way to identify them, count them and idenitify their position? A naive way would be to split the image in sub-images and identify them on each sub-image, and then recompose the image.
I’ve seen this approach: https://www.cv-foundation.org/openaccess/content_cvpr_2015/papers/Oquab_Is_Object_Localization_2015_CVPR_paper.pdf but I am curious if there is a fast.ai way to do that.
F

1 Like
2

In lesson 7 from last year there’s an example of Class Activation Maps (CAM). Here is a paper with more details and examples.

Hope this helps!

1 Like
3

Thanks!