Some problem about training cfn_jps.caffemodel

[xuyanging]

I first use generate_jps_dataset.bin to convert usual lmdb dataset into jps.lmdb
and then product training using train_val_cfn_jps.prototxt.
but i come accross a problem as follows:
...
I1106 20:58:08.315138 7442 net.cpp:159] Top shape: 50 256 1 1 (12800)
I1106 20:58:08.315145 7442 net.cpp:167] Memory required for data: 6323800
I1106 20:58:08.315148 7442 layer_factory.hpp:77] Creating layer pool2_s1
I1106 20:58:08.315167 7442 net.cpp:108] Creating Layer pool2_s1
I1106 20:58:08.315176 7442 net.cpp:456] pool2_s1 <- conv2_s1
I1106 20:58:08.315239 7442 net.cpp:413] pool2_s1 -> pool2_s1
*** Aborted at 1509973088 (unix time) try "date -d @1509973088" if you are using GNU date ***
PC: @ 0x7f12bee65e9e caffe::Blob<>::Reshape()
*** SIGFPE (@0x7f12bee65e9e) received by PID 7442 (TID 0x7f12bf609a80) from PID 18446744072617352862; stack trace: ***
@ 0x7f12bd20e4b0 (unknown)
@ 0x7f12bee65e9e caffe::Blob<>::Reshape()
@ 0x7f12bee662ba caffe::Blob<>::Reshape()
@ 0x7f12bed89712 caffe::PoolingLayer<>::Reshape()
@ 0x7f12beea0df4 caffe::Net<>::Init()
@ 0x7f12beea2811 caffe::Net<>::Net()
@ 0x7f12bee53a32 caffe::Solver<>::InitTestNets()
@ 0x7f12bee543c5 caffe::Solver<>::Init()
@ 0x7f12bee546d9 caffe::Solver<>::Solver()
@ 0x7f12bee7eeb3 caffe::Creator_SGDSolver<>()
@ 0x40a588 train()
@ 0x4074a0 main
@ 0x7f12bd1f9830 __libc_start_main
@ 0x407c09 _start
@ 0x0 (unknown)
Floating point exception (core dumped)

could you give some suggestions? When you produce jps.lmdb, is there any tricks or datails?