I passed the 3 exercises of this assignment and the model got an accuracy of 0.977 with really amazing predicted masks. But the grader gives me 66/100.
Of course, there must be a reason, but could there be a hint as to what should have been done better?.
Did the grader give you any more details when you click “Show Grader Output”? Sometimes it will at least tell you which function(s) it did not accept.
I did not realize that there was a feedback from the grader. With it I quickly found the error.
In fact it was the cause number 3 in the list given in @sjfischer answer:
3. Do not hardcode the size or dimensions of any objec
I would suggest not to put the number of classes in the last instruction of the model (conv10=…) as you could think from a quick reading of the first sentence but the appropriate name in the header of the function.