Andrew Hundt

Wow, fantastic! Thank you! I've just started grad school classes up again so time is a bit dear at the moment, but I have this in my project issue tracker...

I took a look at the code and ran the demo example, it is very impressive! By manually mousing around I can't get it to exceed ~20% cpu even if...

For two classes you should probably be using sigmoid, which just gives a score between 0 and 1. As for the specific error... I'm not sure I've trained on two...

@IvanEz great! I'm glad that worked. :-)

probably not enough data. Also, if one class such as background dominates class weighting may help escape the trap, assuming there is enough data.

@IvanEz I haven't used class weight myself yet so I can't comment on that accurately.

It works fairly well and I merge any good pull requests people submit, plus I've been adapting bits and pieces into [keras-contrib](https://github.com/farizrahman4u/keras-contrib) as I can. That said, to my knowledge...

If I recall correctly I believe it does train on variable sizes. The upsampling is just because some of the networks downsample by a factor of 2 or 4, so...

there are also some cmake changes to gfx, I'm not sure those are actually captured by this.

the scripts should generate those files if you follow the readme instructions