Nikos Pitsillos

I apologise is this is not the correct place but I didn't find anything elsewhere. Why are rewards reversed and why do we append R to the values at the...

This seems like a nice feature to have! I would need this too if it's possible.

Maybe this is a hack to this problem but for me this works. For me it was also giving some weird errors saying something about _get_distribution_strategy. I uncommented the callback...

@JacopoMilone Initially let me state this may not work for you but one can migrate mask rcnn to tf2. Now in my case I was extending the architecture for some...

Same errors? And did you try uncommenting the callback? I'll also check my code to see what else I changed.

Are you using the model as obtained from this repo the model.py I mean? If no please add your code here so I can check with what I have done...

It is a bit cumbersome here to read that. Can you add the file in a Github gist? I'll try to see what the problem is but it will take...

Can you please show a list of your environment packages?

I can't see any major differences in our code but I'll keep checking.

Honestly the only thing I can say is to subclass the Layer class like so: ``` class AnchorsLayer(KE.Layer): def __init__(self, anchors, **kwargs): super(AnchorsLayer, self).__init__(**kwargs) self.anchors = anchors def call(self, image_placeholder):...