Unet: Problem with Model Accuracy

Hi, I passed all the in-notebook grading and got 100/100 for submission, but I am wondering if there is something wrong in my model since my accuracy graph looks like this:


Curious if anyone else has this problem.

I have encountered accuracy fluctuations while training on other problems. This is where you can use EarlyStoppingCallback. That said, here’s an accuracy plot based on the latest version of the notebook.


I have experienced accuracy fluctuations as well, the first time it happened it was for epochs 39 and 40 reducing accuracy significantly in the end. After rebooting the server and restarting the kernel it looks fine

The course doesn’t use tensorflow 2.9.

The only way for us to compare results is by setting the same tf.random.set_seed which may not always be close.

Keep an eye on this starting tensorflow 2.9.