Raspberry Pi 4で YOLOv4を動かす

• Raspberry Pi 4 4GB

• logicool C270N

• 2021-01-11-raspios-buster-armhf-full

パッケージの更新とスワップ領域の設定

sudo apt update
sudo apt upgrade

sudo nano /etc/dphys-swapfile

/etc/dphys-swapfile

CONF_SWAPSIZE=4096
CONF_MAXSWAP=4096

sudo /etc/init.d/dphys-swapfile stop
sudo /etc/init.d/dphys-swapfile start

各種パッケージのインストール

sudo apt install python-opencv
sudo apt install libopencv-dev
sudo apt install libopm-dev
sudo pip3 install opencv-python

YOLOv4のmake

git clone https://github.com/AlexeyAB/darknet
cd darknet/
nano Makefile

Makefile

 OPENCV=1
 OPENMP=1
 LIBSO=1
 USE_CPP=1

make -j4

ウェイトファイルのダウンロードと動作確認

wget https://github.com/AlexeyAB/darknet/releases/download/darknet_yolo_v4_pre/yolov4-tiny.weights

静止画の検出
./darknet detector test cfg/coco.data cfg/yolov4-tiny.cfg yolov4-tiny.weights data/dog.jpg

cfg/yolov4-tiny.cfgにて入力画像サイズ（width、height）を変更した時の推論時間
416x416 data/dog.jpg: Predicted in 5673.473000 milli-seconds.
256x256 data/dog.jpg: Predicted in 2625.850000 milli-seconds.
128x128 data/dog.jpg: Predicted in 516.762000 milli-seconds.

128x128 Overclocking:1800MHz
data/dog.jpg: Predicted in 436.980000 milli-seconds.

動画の検出
./darknet detector demo cfg/coco.data cfg/yolov4-tiny.cfg yolov4-tiny.weights -c 0

FPS:1.1 AVG_FPS:1.1