I have implemented Image Segmentation using transfer learning in the U-net architecture with ResNet-34 dataset..
The highest accuracy achieved with this implementation was 69.45% while the highest accuracy ever achieved by anyone on this dataset is 74%.
Caution: The length of this data is terrbly large and hence you'd not be able to view it entirely from here. I suggest you should clone it somewhere in you local computer and try to view it over there.
Below are the results after the training.
