MobileNet-SSD-TensorRT

To accelerate mobileNet-ssd with tensorRT

TensorRT-Mobilenet-SSD can run 50fps on jetson tx2

Requierments:

1.tensorRT4

2.cudnn7

3.opencv

Run:

cmake .
make
./build/bin/mobileNet

Reference:

https://github.com/saikumarGadde/tensorrt-ssd-easy

I replaced depthwise with group_conv,because group_conv has been optimized in cudnn7

I retrianed mobileNet-SSD,my number of classfication is 5

TODO:

To save serialized model
To solve the bug of getting different result with same input
The bottleneck of time cost lies in the decoding of pictures. "imread" cost too much ,to resolve it.
To modify the architecture, decrease the time cost

If want to decrease the time cost of "imread",you could rebuild OpenCV[https://github.com/jetsonhacks/buildOpenCVTX2]

Added producer-consumer

The bug has been fixed

Name		Name	Last commit message	Last commit date
Latest commit History 16 Commits
model		model
testPic		testPic
testVideo		testVideo
util		util
.DS_Store		.DS_Store
CMakeLists.txt		CMakeLists.txt
README.md		README.md
common.cpp		common.cpp
common.h		common.h
cudaUtility.h		cudaUtility.h
imageBuffer.h		imageBuffer.h
kernel.cu		kernel.cu
main.cpp		main.cpp
mathFunctions.cpp		mathFunctions.cpp
mathFunctions.cu		mathFunctions.cu
mathFunctions.h		mathFunctions.h
mobileNet.cbp		mobileNet.cbp
pluginImplement.cpp		pluginImplement.cpp
pluginImplement.h		pluginImplement.h
tensorNet.cpp		tensorNet.cpp
tensorNet.h		tensorNet.h

jamessmith90/MobileNet-SSD-TensorRT