thank you for the great article , I had horrible time with this challenge , started too late , and was clueless what to do most of the time..
I have some stupid questions :
1- how did you figure out what layers/variables to load from DeepLab ? I mean how to do that with any model I chose to build my project on ?
2- you said “ I tried further augmentation but it didn’t prove useful “ , how did you decide that ? you are training for 300 epoches ! how to know if augmentation works or useless ?
3- you tried many models , but I was wandering : do you use same learning_rate / keep_prob / variable_initializer / with all models ? how do you it is the model that is not improving the result , not there parameters (because I stuck my self in trying diffrent variations of learning rates and keep probabilities .