One of the most common approaches to Transfer Learning is using a model which has been pre-trained to classify ImageNet.
I’m working on a fine-grained image segmentation problem. I’m curious if anyone has experimented with pre-training a model which performs image segmentation on some large, open image segmentation datasets?
If yes, did it out-perform the classic pre-trained ImageNet model?
Also, If you can share any useful resources on this topic, that would be very helpful.
All the best,