I created a VGG based U-Net in order to perform image segmentation task on yeast cells images obtained by a microscope.
There are a couple of problems with the data:
- There is inhomogeneity in the amount of yeast in the images. 1 image can have hundred of yeast cells while others can have less the one hundred.
- The GT segmentation map is also incomplete, and some of the cells are not labeled.
All in all the model, given the above problem, is able to learn in some manner. My problem is that the segmentation maps seem incomplete.
For example:
My loss function contains BCE, I was wondering if there is a way to force the model to create a 'fuller' segmentation maps. Something like using Random fields of some sort. Or maybe to enhance my loss function to overcome the above-mentioned problems.
I wish to stay in the domain of simple architectures rather than using more sophisticated ones such as RCNN.
Would appreciate any suggestions
