# Xavier、tx2安裝

壓縮檔裡面附上Xavier所用環境的Dockerfile，裡面的腳本(Jetson.Dockerfile, install_protobuf.sh)有使用"皮愛皮"關鍵字需要修改

編譯Dockerfile成DockerImage的指令
docker build -t Image名稱 -f Jetson.Dockerfile .

內部已經自動下載好darknet和torch2trt，由於Jetson在建立環境的時候無法讀到cuda, cudnn，因此進入container後需要手動編譯darknet

### Jetson.Dockerfile 內容

#ngc jetson pytorch1.4
FROM nvcr.io/nvidia/l4t-pytorch:r32.4.2-pth1.4-py3

RUN apt-get update && apt-get install -y cmake unzip

#Install onnx dependency
COPY ./install_protobuf-3.8.0.sh /install_protobuf-3.8.0.sh
RUN 皮愛皮3 install --upgrade 皮愛皮 && \
bash /install_protobuf-3.8.0.sh && \
rm -f /install_protobuf-3.8.0.sh && \
rm -rf /root/src

#Install python packages
COPY ./jetson_requirements.txt /requirements.txt
RUN  皮愛皮3 install -r /requirements.txt && \
rm -f /requirements.txt


#For chinese comments
ENV LANG C.UTF-8

#Install torch2trt
RUN git clone https://github.com/NVIDIA-AI-IOT/torch2trt /git/torch2trt
WORKDIR /git/torch2trt
#I can't build torch2trt when image building, but I can build it when it is a container.
#RUN python3 setup.py install


#---Install darknet---
#--Clone darknet to /git/darknet
RUN apt-get update && \
apt-get install -y libopencv-dev \
build-essential \
pkg-config \
cmake && \
git clone https://github.com/AlexeyAB/darknet.git /git/darknet

#--Jump to darknet WORKERDIR
WORKDIR /git/darknet

#--Checkout to specialVersion(Yolov4) and compiler
RUN git checkout 38a164bc
#
#make OPENCV=1 GPU=1 AVX=0 OPENMP=1 CUDNN=1 CUDNN_HALF=1 OPENMP=1 LIBSO=1 -j $(nproc)
#RUN chmod +x ./darknet

ENV DARKNET_PATH=/git/darknet
#Add darknet.py to PYTHONPATH
ENV PYTHONPATH=/git/darknet:$PYTHONPATH

### install_protobuf-3.8.0.sh
**install_protobuf-3.8.0.sh -> https://github.com/d246810g2000/trt_yolv4-tiny

https://d246810g2000.medium.com/nvidia-jetson-nano-%E4%BD%BF%E7%94%A8-yolov4-tiny-%E9%80%B2%E8%A1%8C%E4%BA%BA%E8%87%89%E5%81%B5%E6%B8%AC-31944262e8f8

### jetson_requirements.txt

easydict==1.9
opencv-python==4.4.0.42
tqdm==4.48.2
Flask==1.1.2
requests
notebook
matplotlib
ffmpeg-python
onnx==1.4.1
pycuda>=2017.1.1

### 環境清單.txt

apt-get安裝
cmake, unzip, git, libopencv-dev \
build-essential \
pkg-config \
cmake


darknet(已經clone在/git/darknet，須手動編譯):
git clone https://github.com/AlexeyAB/darknet.git(38a164bc)


torch2trt:(Container環境內的torch2trt預設還沒安裝，要安裝請進到/git/torch2trt，執行 python3 setup.py install)
git clone https://github.com/NVIDIA-AI-IOT/torch2trt



請先安裝install_protobuf.sh(超過30分鐘)
才可以安裝onnx

皮愛皮安裝
torch==1.4.0
torchvision==0.5.0



easydict==1.9
opencv-python==4.4.0.42
tqdm==4.48.2
Flask==1.1.2
requests
notebook
matplotlib
ffmpeg-python
onnx==1.4.1
pycuda>=2017.1.1