Training loss NaN
Hello Andrew,
first of all, very nice work, congrats! I was testing the Hierarchical CADNET with your already generated batches and this worked well.
1st Sub-Issue: When I was generating batches myself from the STEP-data, I also got the "NoneType Object has no attribute Nodes" (currently I use pythonocc-core=7.4.1 occt=7.4.0, tried also pythonocc-core=7.5.1 and occt=7.5.1, where I get this message only for the model 42635 "test"-batches). I would be grateful, if you would tell me, if this is a problem (I assume, this model is simply not taken into account for the test?).
2nd (higher priority) Sub-Issue: I take these "self-generated" .h5 files (based on your data-set) and try to use them for Hierarchical CADNET, but soon (already in the 1st Epoch, sometimes already for the first batch, sometimes after 5-6 batches) I get a NaN Training loss. Could you with your experience have a suggestion how to find what the problem is? I was using the given environment.
A remark: I was downloading the source files, could this matter?
Thanks a lot and best regards, Jovana