leolee99 / yolov4_siamrpn Goto Github PK

• CUDA 10.2
• OpenCV 4.1.1 with contrib(KCF need)
• TensorRT 7.1.3.0
• cuDNN 8.0.0
• Python3.6
• pip3 (make sure it's the new version:20.3.3)
• pycuda (run pip3 install pycuda)
• Numpy (run pip3 install numpy)
• pytorch 1.6.0 (without Anaconda)
• opencv-python 4.4.0.46
• tqdm
• pyyaml
• matplotlib
• colorama
• cython
• tensorboardX

sudo apt-get purge libopencv* # delete old version

cd ~

bash install_opencv4.1.1_jetson.sh # download the source and install

In file included from /home/esraa/opencv_contrib/modules/cudaarithm/src/cuda/normalize.cu:53:0:
/home/esraa/opencv/modules/core/include/opencv2/core/private.cuda.hpp:75:0: warning: "NPP_VERSION" redefined
 #  define NPP_VERSION (NPP_VERSION_MAJOR * 1000 + NPP_VERSION_MINOR * 100 + NPP_VERSION_BUILD)

Just use ifndef and define solve it

solution

~/opencv_contrib/modules/xfeatures2d/src/boostdesc.cpp:673:20: fatal error: boostdesc_bgm.i: No such file or directory

because of the network problem,i can't download those files

solution

pip3 install numpy pip3 install pytorch pip3 install opencv-python

pip3 install pyyaml yacs tqdm colorama matplotlib cython tensorboardX

python setup.py build_ext --inplace

There are always some errors when we install pytorch, opencv-python and some other libs on aarch64 architecture by pip. Then you can download source from pytorch.org or pypi.org, excute the following instructions on terminal to install them.

python3 setup.py build
python3 setup.py install

export PYTHONPATH=/home/nvidia/yolo_siamrpn++:$PYTHONPATH #Your project path

Download models in PySOT Model Zoo and put the model.pth in the correct directory in experiments.

python3 trt_demo.py --config experiments/siamrpn_r50_l234_dwxcorr/config.yaml --snapshot experiments/siamrpn_r50_l234_dwxcorr/model.pth --video_name video/2.flv

python3 trt_demo.py --config experiments/siamrpn_mobilev2_l234_dwxcorr/config.yaml --snapshot experiments/siamrpn_mobilev2_l234_dwxcorr/model.pth --video_name video/2.flv

Jetson Packages install list

miniforge(better than archiconda)