Lessons 1 and 2 show how to make and run a classifier which sorts images into three categories (black bear, brown bear, teddy bear). If I instead wanted to make a classifier for whether or not an image is something, such as a banana, how would I go about doing that? Would I just make two categories, banana and not banana, and find images of bananas and images of other objects? Or is there a simpler way to do this, without a category of all other images? Any help would be much appreciated.
Have a look at the ‘unknown labels’ notebook by @muellerzr and it should guide you in the right direction.
2 Likes
(Resubmitted to make sure I didn’t mess up the formatting).
Thanks! I read the notebook, and it does seem to provide a fair solution, but I’m still wondering: What would happen if I only had one category and wanted to recognize whether or not something fit in that category? Would the machine be able to make any meaningful adjustments if all the pictures in the training set were one specific category, or would it just learn to mark any image it saw as that part of the category (since the training set is only images from that category)?