Jetson nanoでyolov3,yolov3-tinyを動かすメモ #JetsonNano

kerasのインストール

sudo apt install libatlas-base-dev gfortran
sudo pip3 install scipy
sudo pip3 keras

keras-yolo3リポジトリ

git clone https://github.com/qqwweee/keras-yolo3

weights

YOLOv3

wget https://pjreddie.com/media/files/yolov3.weights

YOLOv3-tiny

wget https://pjreddie.com/media/files/yolov3-tiny.weights

weightsのコンバート

YOLOv3

python3 convert.py yolov3.cfg yolov3.weights model_data/yolo.h5

YOLOv3-tiny

python3 convert.py yolov3-tiny.cfg yolov3-tiny.weights model_data/yolo-tiny.h5

カメラ入力部分の修正※Raspberry Pi Camera Moduleの場合

yolo.py

def detect_video(yolo, video_path, output_path=""):
    import cv2
    GST_STR = 'nvarguscamerasrc \
    ! video/x-raw(memory:NVMM), width=3280, height=2464, format=(string)NV12, framerate=(fraction)30/1 \
    ! nvvidconv ! video/x-raw, width=(int)416, height=(int)416, format=(string)BGRx \
    ! videoconvert \
    ! appsink'
    WINDOW_NAME = 'Camera Test'
    vid = cv2.VideoCapture(GST_STR, cv2.CAP_GSTREAMER)
#    if video_path.isdigit():
#        video_path = int(video_path)
#    vid = cv2.VideoCapture(video_path)
    if not vid.isOpened():

※USBカメラの場合はvid = cv2.VideoCapture(0)

parser文字列の修正（３ヶ所）

yolo_video.py

#    '--model', type=str,
    '--model_path', type=str,

#    '--anchors', type=str,
    '--anchors_path', type=str,

#    '--classes', type=str,
    '--classes_path', type=str,

実行　※GUI環境

YOLOv3 FPS=1　※重くてvideo出力がハングアップ状態になることがある

python3 yolo_video.py

YOLOv3-tiny FPS=4

python3 yolo_video.py --model_path model_data/yolo-tiny.h5 --anchors_path model_data/tiny_yolo_anchors.txt

その他

「importerror: The _imagingft C module is no installed」の対処

sudo apt-get install libfreetype6
sudo apt-get install libfreetype6-dev
sudo pip3 uninstall Pillow
sudo pip3 install Pillow

Jetson nanoでyolov3,yolov3-tinyを動かすメモ