caffe model convert to tensorrt error

python: caffe/caffeParser.cpp:104: virtual nvinfer1::DimsHW CaffeParserPoolingDimsCallback::compute(nvinfer1::DimsHW, nvinfer1::DimsHW, nvinfer1::DimsHW, nvinfer1::DimsHW, nvinfer1::DimsHW, const char*): Assertion `input.h() + 2 * padding.h() >= kernel.h()’ failed.
Aborted (core dumped)

i have query this issue with google, and found that this issue is a bug in tensorrt3, and the official said will fix in tensorrt4.0. meanwhile it still error in tensorrt 4.0 with ubuntu16.04, cuda9.0