Segmentation error in coffee-based neural network convolutional model

I am following a human segmentation tutorial that trains a model FCN-8s

caffe

for a semantic segmentation problem using Pascal VOC-12, in which out of 21 classes, one class has a personality, and furthermore, they trained a similar model again, but for 1 class, which is a human in portrait mode and the images in the sample database are 800 X 600. But when I tried to prepare 360 ​​X 640 person images for landscape mode, I get segmentation error while setting fc6 convolutional layer, the error log is in the bottom link as it is too big to post here. error log

training script:

protobuf file for FCN-8 model:

solution file for the model used at the top of the FCN-8:

send the protobuf file for the model used at the top of the FCN-8: