Tensorrt on pva-faster-rcnn


I am trying to optimize pva-faster-rcnn using tensorrt. pva-faster-rcnn is similar to faster-rcnn except some extra layers. I have tried using sampleFasterRcnn plugin for that and i have changed the parameters accordingly. I have also modified the prototxt similar to that to fasterRcnnIplugin.prototxt. But i am getting segmentation fault. Kindly help me in this. I have compared both the prototxt of fasterrcnn and pva-faster-rcnn, i have made modifications according to fsater-rcnn to inlcude plugin module. Thanks.

Hello, can you provide details on the platforms you are using?

Linux distro and version
GPU type
nvidia driver version
CUDA version
CUDNN version
Python version [if using python]
Tensorflow version
TensorRT version

And maybe a trace callback if available.

Hi, here are the details:

Ubuntu 16.04
GeForce GTX 1080
nvidia Driver Version: 396.37
Not using python, i am using c++.
Not tensorflow, i am using caffe.
TensorRT -

I debugged with gdb, i got the following message.

"Starting program: /home/alien1/Downloads/TensorRT-
[Thread debugging using libthread_db enabled]
Using host libthread_db library “/lib/x86_64-linux-gnu/libthread_db.so.1”.
[New Thread 0x7fffd1003700 (LWP 27763)]
[New Thread 0x7fffd0802700 (LWP 27764)]
[New Thread 0x7fffbff7e700 (LWP 27765)]
Begin parsing model…

Thread 1 “sample_fasterRC” received signal SIGSEGV, Segmentation fault.
0x00007ffff1cee410 in parseInnerProduct(nvinfer1::INetworkDefinition&, ditcaffe::LayerParameter const&, CaffeWeightFactory&, BlobNameToTensor&) () from /usr/lib/x86_64-linux-gnu/libnvparsers.so.4.1.2
(gdb) "

It is failing in parsing stage which parses the prototxt and caffemodel.
You can refer this repo for pva-models: https://github.com/sanghoon/pva-faster-rcnn

Hope this helps you figure out.


Thank you for the information. Will keep you updated on what we find.

Hi Zhenyih,

Have you found any clue?

hello k.sambasivarao13

Can you please provide either
(1) the modified sampleFasterRCNN.cpp which exhibits the failure, or

(2) the modified prototxt file and trtexec/giexec command the used to reproduce this crash?

It will be hard for us to do any meaningful debugging without a local repro.


Thanx for the response, how can i attach here. Could you share your emailid so that i can share on google drive.



you can click on my name and private message me.


I meet this issue recently, the platform and library version as the same as k.sambasivarao13
Did you slove this issue? Could you provide resolution about use tensorRT on pva-faster-rcnn?