Finetuning FCN on a new dataset

[sgs]

Hi,

I want to finetune FCN8s on my own dataset. For that i've created the matrix for pixel labels for each of my image. When I try to train the model this is the error I get. 

 

Check failed: outer_num_ * inner_num_ == bottom[1]->count() (120000 vs. 360000) Number of labels must match number of predictions; e.g., if softmax axis == 1 and prediction shape is (N, C, H, W), label count (number of labels) must be N*H*W, with integer values in {0, 1, ..., C-1}.

My images are 300x400. Since the labels are of (1,1,300,400) and images are (1,3,300,400) the prediction count doesn't match with the input pixel count, I get this part. But how to take care of this issue? Which codes should I pay attention to? I am new to caffe so it'd be great if someone can give some pointers. Thanks.