There is no plan to train DPNs on Caffe.
But you can find some converted model at: soeaver/caffe-model
Also, you can convert the trained mxnet model by using model converters: ysh329/deep-learning-model-convertor

Yes，I know this converters, thanks for your help.
another question: the score.py is running wrong ,the val.log is empty, there is no any result.

Do you have VeiXin , can I add you?

"--network dpn-92" in the run_val.sh , show the wrong "Segmentation fault(core dumped)", how can I do?

Solved