Check failed: num_priors_ * num_classes_ == bottom[1]->channels()

[gongdalinux]

I use the command './train.sh' to train my datasets. The issue is

F0117 21:34:50.957195 1386 multibox_loss_layer.cpp:141] Check failed: num_priors_ * num_classes_ == bottom[1]->channels() (44091 vs. 41925) Number of priors must match number of confidence predictions.
*** Check failure stack trace: ***
@ 0x7f11efa7d5cd google::LogMessage::Fail()
@ 0x7f11efa7f433 google::LogMessage::SendToLog()
@ 0x7f11efa7d15b google::LogMessage::Flush()
@ 0x7f11efa7fe1e google::LogMessageFatal::~LogMessageFatal()
@ 0x7f11f0323a02 caffe::MultiBoxLossLayer<>::Reshape()
@ 0x7f11f0388933 caffe::Net<>::Init()
@ 0x7f11f038a161 caffe::Net<>::Net()
@ 0x7f11f01be78a caffe::Solver<>::InitTrainNet()
@ 0x7f11f01bfa87 caffe::Solver<>::Init()
@ 0x7f11f01bfe2a caffe::Solver<>::Solver()
@ 0x7f11f03a88e9 caffe::Creator_RMSPropSolver<>()
@ 0x40afd9 train()
@ 0x4077e8 main
@ 0x7f11ee213830 __libc_start_main
@ 0x4080b9 _start
@ (nil) (unknown)
Aborted (core dumped)

num_classes = 23; Should I change the last layer mbox_loss? More details will be appreciated
ubuntu16.0+cuda9.0
TKS!