Tensorrt on pva-faster-rcnn

AI & Data Science Deep Learning (Training & Inference) TensorRT
1

Hi,

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.

2

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.

3

Hi, here are the details:

Ubuntu 16.04
GeForce GTX 1080
nvidia Driver Version: 396.37
cuda-9.2
cudnn-7.1.4
Not using python, i am using c++.
Not tensorflow, i am using caffe.
TensorRT - 4.0.1.6

I debugged with gdb, i got the following message.

"Starting program: /home/alien1/Downloads/TensorRT-4.0.1.6/targets/x86_64-linux-gnu/bin/sample_fasterRCNN_debug
[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: GitHub - sanghoon/pva-faster-rcnn: Demo code for PVANet

Hope this helps you figure out.

Thanx

4

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

5

Hi Zhenyih,

Have you found any clue?

6

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.

7

Hi,

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

Thanx

8

Hi,

you can click on my name and private message me.

thanks

9

Hi NVES,
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?

Thanks