[Resolution expected] I'm not mad (0.11.11)

(Doth he protest too much?)

I rebuilt the seeing in the dark model and have noted that when opened in 0.11.11 it is missing all the links and there’s that “NoneType” spec_class issue…

I thought I’d look closer and exported the code to ipynb. I’ve attached the result. Seeing in the dark rebuilt.zip (11.6 KB)

Key takeaway (apart from “Graph” still missing - which I already knew about from @robertl), I was surprised that there are internal inconsistencies: Is all of this perhaps attributable to that single “NoneType” error noted above?

  • Jupyter: “NameError: name ‘DeepLearningConv_Convolution_14_1’ is not definedDeepLearningConv_Convolution_14_1”
  • There isn’t a class for DeepLearningConv_Convolution_15_1 either AFAICT

In fact all these have to be commented out to get the layer_classes dict to work

 #   'DeepLearningConv_Convolution_14_1': DeepLearningConv_Convolution_14_1,
 #   'DeepLearningConv_Convolution_15_1': DeepLearningConv_Convolution_15_1,
 #   'DeepLearningDeconv_Deconvolution_4_1': DeepLearningDeconv_Deconvolution_4_1,
 #   'DeepLearningDeconv_Deconvolution_5_1': DeepLearningDeconv_Deconvolution_5_1,
 #   'DeepLearningDeconv_Deconvolution_6_1': DeepLearningDeconv_Deconvolution_6_1,
 #   'MathMerge_Merge_5_1': MathMerge_Merge_5_1,
 #   'TrainNormal_Classification_1_1': TrainNormal_Classification_1_1,
 #   'DeepLearningDeconv_Deconvolution_2_1': DeepLearningDeconv_Deconvolution_2_1,

Yet e.g. Convolution_14_1 is most definitely there (even if you can’t see the links, you will see the edges in the ipynb)

NB I compared it visually to the original Seeing in the dark model and it looks the same… but where are the deconvolutions? They are listed in the components - albeit without a colour code…

PS Here’s long output from the cmd window*

I don’t actually know which model(s) are being referred to, but you will find “NoneType” errors and persistent, repeated issues with the same Deconvolution_1_1 (but not only that layer)

Hope it’s informative
V0-11-11 NoneType errors.txt (128.3 KB)

Hi @JulianSMoore,
Thanks for the great logs, we will take a look!
I do think it may be something crashing when using an old model in 0.11.11, but we will verify more precisely.

If I understood @robertl correctly, the issue here is that the Deconvolution component was omitted from 0.11.11 - it will return as a setting of Convolution soon, but until then, models built in previous versions using deconvolution will no longer work and notebook code will not contain the classes required for the components on such diagrams.

I don’t think it will be a long wait for The Return of the Deconvolution…