RSly

thanks @ethantang95 So it seems that the number of classes are hard coded to be 2 in this implementation. any simple way to make it multi-class unet? I see in...

it seems it is solved by re-writing top[0].data[0]=np.sum(dice)/float(bottom[0].data.shape[0]) I still have some problems... I will investigate and let you know

here is more details regarding the batch_size problem: figure 1. with batch_size : 1 the network learns nicely, and achieves 90% accuracy  figure 2. with batch_size : 8, it...

@getta, thanks a lot for the suggestion and the links. I tried the 0.7 decrease. it actually helps, so using this trick, the results with batch_size:1 and batch_size 10 are...