Batch size built into the model [tf2 branch]

[bklooste]

The batch size if built into the model , this creates quite a few issues and can fail when dealing with different training sizes or filtering data .

[CoffeeStraw]

Have you managed to solve that?

[essamgoda]

I have Same error.

[bklooste]

Its a fair bit of work due to the way it was done .. almost need to redo .

[CoffeeStraw]

I think I've managed to make this work... Unfortunately, I've organized my codebase differently from Xifeng implementation (even if my work is based on its implementation) and I will not be able to prepare a PR for this project in a while. For those interested, here's my work:

https://github.com/CoffeeStraw/CapsNet-Knowledge-Extractor/blob/master/capsnet_trainer/_share/capslayers.py