Black output

[Sam813]

Hi,
I have tried to train the CRFasRNN with my data which are CT medical images. My images consist of two class of background and tumors.
I used the training prototext in this repo and seems the training is going on well, but when I check the prediction of the network on testing images, all are in black.

I read that I have a problem in weight initialization of the convolution/ deconvolution layer, but I don't know how to fix it.
Can anyone please help me on this?

[KleinYuan]

@Sam813 refer to this reply

[Sam813]

@KleinYuan Thanks for the fast response but sorry I don't get what you mean? My loss is reducing in during the training. The problem is that all predictions of the network are in black.
I was wondering what is the problem that all the outputs are black. In
here they mentioned possibly I have not initialized the weights for deconvolution layers correctly.
So may I know how should i initialize it.
below is the way that i did it
layer {
name: "score2-new"
type: "Deconvolution"
bottom: "score"
top: "score2-new"
param {
lr_mult: 1
}
convolution_param {
num_output: 2 #21
kernel_size: 4
stride: 2
weight_filler: { type: "xavier"
std: 0.1
}
bias_filler {
type: "constant"
value: 0.2 }
}
}
So as you see, i changed the number of outputs to 2 and initialize the weights by xavier method.

I have included my train prototext in case it can help.
Looking forward to your assist
Thanks

Train_prototext.txt