guofenggitlearning/LightTrack-rknn

RKNN version demo of [CVPR21] LightTrack: Finding Lightweight Neural Network for Object Tracking via One-Shot Architecture Search

LightTrack: Finding Lightweight Neural Networks for Object Tracking via One-Shot Architecture Search (rknn)

The official implementation by pytorch:

https://github.com/researchmm/LightTrack

Run with pre-compiled version

Download pre-release(rv1126/rv1109)

https://github.com/Z-Xiong/LightTrack-rknn/releases/tag/rv1109%2Frv1126

connect your development board

The model only supports rv1109 and rv1126.

install to rv1126/rv1109

$ cd LightTrack-rknn
$ adb push model /userdata
$ adb push 01.mp4 /userdata
$ adb push $PATH_TO_RELEASE/LightTrack /userdata

run by rv1126/rv1109

$ adb shell
# in rk shell
$ cd /userdata
$ ./LightTrack 01.mp4

export out.avi

# in host bash
$ adb pull /userdata/out.avi

Or you can compile it yourself by following the steps below

0. Download lib

链接: https://pan.baidu.com/s/1iDRuForXsz4p1NNQCZ1QIg 密码: fqba

Unzip the zip

$ cp -r lib LightTrack-rknn

1. How to run it?

modify your own CMakeList.txt CROSS_COMPILE_TOOL

set(CROSS_COMPILE_TOOL $YOUR_PATH/rk1109_rk1126_linux/prebuilts/gcc/linux-x86/arm/gcc-arm-8.3-2019.03-x86_64-arm-linux-gnueabihf/bin/arm-linux-gnueabihf)

build

$ mkdir build && cd build
$ cmake .. && make -j
$ make install

connect your development board

The model only supports rv1109 and rv1126.

install to rv1126/rv1109

$ adb push ../install /userdata

run by rv1126/rv1109

$ adb shell
# in rk shell
$ cd /userdata/install/lighttrack_demo
$ ./LightTrack 01.mp4

export out.avi

# in host bash
$ adb pull /userdata/install/lighttrack_demo/out.avi

---

You may encounter build problems, raise an issue.