Matthew Avaylon

With all these changes, there’s areas outside the code base where the issue(s) could be. You also didn’t provide any of the new code so no one can really help...

Also the masks are reshaped but the images are not reshaped when going into the model? @divamgupta

Changing to custom loss and metric still reports accuracy and cross-entropy. It doesn't have errors, it simply just reports accuracy even if you change it.

So a rule of thumb that I've found around is steps per epoch = training size/ batch size. This will ensure using all training data. Changing the batch size is...

I think I see why. Because of the UNet decoder that needs 64*64, you padded the 63*63 in the resnet with the zero on one side. Right?

yes you can

Running this through a profiler, hdmf accounts for roughly 46-50% of the time for importing pynwb. Within the time spent for hdmf, a very small portion is spent on backend,...

And if i downgrade my tf to 1.15 I get "tensorflow.python.framework.errors_impl.NotFoundError: dlopen(/Users/matthewavaylon/crfasrnn_keras/src/cpp/high_dim_filter.so, 6): Library not loaded: @rpath/libtensorflow_framework.2.dylib Referenced from: /Users/matthewavaylon/crfasrnn_keras/src/cpp/high_dim_filter.so Reason: image not found"

Note: I will tackle this in February 2024

@jrhemstad I followed your workflow but I get "Error: Error: Failed to read private key"