Feee
1
I was wondering if there might be an option to YOLOv3 on a Jetson Xavier with a Python wrapper?
I know, there are some wrappers built by users, for example:
https://github.com/madhawav/YOLO3-4-Py
or https://github.com/mosheliv/tensortrt-yolo-python-api
but none of them was possible to install on my Jetson Xavier. Is there anyone who ever made some experience with a python wrapper for yolov3?
Hi LordCalvin, there is a YOLOv3 Python sample included with TensorRT, found at: /usr/src/tensorrt/samples/python/yolov3_onnx
Perhaps you could modify it to fit your needs for a YOLO API.
I use latest opencv release 4.2.0. Opencv dnn module has darknet support with cuda and python interface.
@scipio.it, can you use opencv yolov3 on xavier?
Yes, on xavier I can run 6 yolo networks on two fullhd streams in real time at 30FPS
Just wondering did you manually build the opencv 4.2.0 from source on Xavier? OR
It can be done using pip installer only?
@scipio.it
This is my script for opencv 4.2.0 build, inside a python 3.7 virtualenv as normal user and with dnn CUDA support.
#!/bin/bash
# sudo apt install -y python3.7 python3.7-dev python3.7-venv
# sudo apt install -y libtbb2 libtbb-dev
CUDA="-D CUDA_TOOLKIT_ROOT_DIR=/usr/local/cuda -D WITH_CUDA=ON -D WITH_CUDNN=ON -D CUDA_GENERATION=Auto"
mkdir -p ~/external
cd ~/external
if [ ! -d opencv-4.2.0 ]; then
[ -f 4.2.0.tar.gz ] || wget https://github.com/opencv/opencv/archive/4.2.0.tar.gz
tar zxvf 4.2.0.tar.gz
fi
if [ ! -d opencv_contrib ]; then
git clone https://github.com/opencv/opencv_contrib.git opencv_contrib
fi
mkdir -p opencv-4.2.0/release
cd opencv-4.2.0/release
JOBS=$(getconf _NPROCESSORS_ONLN)
JOBS=$(($JOBS - 1))
cmake \
-D CMAKE_C_COMPILER="/usr/bin/cc" \
-D CMAKE_CXX_COMPILER="/usr/bin/c++" \
${CUDA} \
-D CUDA_NVCC_FLAGS="-D_FORCE_INLINES" \
-D CMAKE_BUILD_TYPE=RELEASE \
-D BUILD_LIST=core,improc,videoio,dnn,python3,cudev,dnn_objdetect,highgui,video \
-D CMAKE_INSTALL_PREFIX=${VIRTUAL_ENV} \
-D OPENCV_EXTRA_MODULES_PATH=../../opencv_contrib/modules \
-D WITH_TBB=ON \
-D ENABLE_PRECOMPILED_HEADERS=OFF \
-D BUILD_PERF_TESTS=OFF \
-D CV_TRACE=OFF \
-D WITH_OPENEXR=OFF \
.. && \
make -j${JOBS} && \
make install/strip
Great and Thanks!! scipio.it