Issue with calling the VanillaCNN nn

[asnokhai]

I have been struggling for a long time with this issue, and I am pretty stuck.

When I set 'model/backbone' to 'resnet', everything functions fine.

However, when I set it to 'vanilla', I get an AttributeError:

I initially thought this was due to my implementation of 'VanillaCNN'. However, when I swap the names of both neural networks, the error remains unchanged:

Here, effectively the 'VanillaCNN' carries the resnet implementation and vice versa. If I now run the code, it works with 'resnet' in the .yaml config, but I get the exact same error as before with 'vanilla'.

Before each run attempt, I cleared all outputs and restarted the python kernel.

How do I fix this?

Thanks in advance!

[yilun-wu]

As the error message suggests, you attempted to squeeze a None object instead of a tensor. From the error trace, it seems like it is the output_angle which is None that get squeezed but failed. Can you check whether that is is the case? And if so, further debug why output_angle is None instead of a pytorch tensor.

[asnokhai]

It is because model(x) returns 'None', but I can't seem to figure out why