train error

[fgxfxpfzzfcc]

Hi,When ｉuse your shufflenet_deploy.prototxt to train my dataset,I encountered the following error:
I1112 22:49:05.710156 18327 net.cpp:454] resx5_concat <- resx5_conv3
I1112 22:49:05.710160 18327 net.cpp:411] resx5_concat -> resx5_concat
Check failed: top_shape[j] == bottom[i]->shape(j) (37 vs. 38) All inputs must have the same shape, except at concat_axis.
the train log is:
Setting up resx5_conv3_scale
net.cpp:157] Top shape: 1 240 38 63 (574560)
net.cpp:411] resx5_match_conv -> resx5_match_conv
net.cpp:150] Setting up resx5_match_conv
net.cpp:157] Top shape: 1 240 37 62 (550560)
so,anyone else who knows how to solver this problem,i will be grateful!