What's the best practice of convert a caffe ssd model who have changed priobox layer?


I have a caffe model and I want to convert it to tensorrt model, It has changed the priobox layer by modify the offical code,so I couldn’t use the original plugin . I have tried to change the cuda code which in offical plugin,and to used in my convert code , I converted it successful, but when I run infer , at last it will issue :

"infer" received signal SIGSEGV, Segmentation fault.
0x00007fb423bc5bf3 in pluginStatus_t allClassNMS_gpu<float, float>(CUstream_st*, int, int, int, int, float, bool, bool, void*, void*, void*, void*, void*, bool) () from /tensorrt/lib/libnvinfer_plugin.so.7

Should I try to use NetworkDefinition API to convert the model ?


TensorRT Version: 7.xx
GPU Type: t4
CUDA Version: 10.2
CUDNN Version: 8.0
Operating System + Version: ubuntu
Python Version (if applicable): 3.6

Hi @bing1zhi2,
Request you to share your model and script so that we can assist you better.