Building MTL models in PyTorch.
Thanks for a wonderful post.
I being into computer vision, wanted to know as per your experience an use case.
As per the current project that I work on, I have 2 tasks to get done...
Let's take a dummy example of quality evaluation of Some big seeds that can be cut in halves.
So first task is that in an image I would like to detect the cutted seeds out of Gullotine (a box used to cut beans) this is necessary because the cuts are not always perfect. So I would like to detect seed n no seed.
Second task is to classify the cutted halves on based of quality for ex. Good, Rotten, Raw.
I have labbeled bean half data for both detect n classification stage so no need to worry there.
Is there a way to use or you might have an idea that I can use multi task in such bits.
Will be glad to hear back