I am new to the whole realm of machine learning, but I do have some prior experience with AWS' Rekognition. Within Rekognition, you're able to custom label different sections within your images, rather than just the entire image as a whole. I was looking to do something similar within Tensorflow, but despite looking throughout their documentation and searching on SO I've not been able to find anything.
Is this just not doable in Tensorflow, or am I just missing something? If I train a model on separately modeled images, can it pick all of the occurrences of these in an image - Eg. an image with 5 people, all 5 are separately detected?
If there is a better library that is more capable of this, I would love to know - or if there's a work around that I could use to implement this myself I am willing to try it.
It is possible to classify different areas of images in tensorflow using its object detection api. See tensorflow object detection api
You can work through the examples there, they also offer pretrained models to run examples with.
Pytorch also offers multibox object detection see pytorch example, just use whatever framework suits you more.