本專案以 YOLOV8 加上 QT6 視窗程式，即時偵測照片中的物件。請先下載 ui_mainwindow.ui，置於專案下的 ui 目錄。

安裝套件

請先安裝如下套件

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118 --no-cache-dir
pip install pyqt6 ultralytics opencv-python

ModelThread

載入模型可能需要一段時間，所以使用新執行緒載入。

from PyQt6.QtCore import QThread, pyqtSignal
from ultralytics import YOLO

class ModelThread(QThread):
    callback = pyqtSignal(object)
    def __init__(self, parent=None):
        super().__init__(parent)
    def run(self):
        model=YOLO("./yolov8n.pt")
        self.callback.emit(model)

PictureThread

此段是載入檔案中的圖片，如果檔案很多，載入時間也會很久，所以也交由新執行緒載入。

import os
from PyQt6.QtCore import QThread, pyqtSignal
from PyQt6.QtGui import QPixmap

class PictureThread(QThread):
    callback=pyqtSignal(object)
    def __init__(self, path, parent=None):
        super().__init__(parent)
        self.path=path
        self.runFlag=True
    def run(self):
        ls = os.listdir(self.path)
        files = []
        for l in ls:
            ll = l.lower()
            if ll.endswith('.jpg') or ll.endswith('.png'):
                files.append(os.path.join(self.path, l))
        index=0
        total=len(files)
        while index < total and self.runFlag:
            pix = QPixmap(files[index])
            pix = pix.scaled(400, 300)
            pix.tag=files[index]
            self.callback.emit(pix)
            index+=1
            QThread.msleep(10)#要停一下, 否則 UI 無法立即顯示

DetectThread

此段為辨識圖片代碼。

import platform
from PIL import Image, ImageFont, ImageDraw
from PyQt6.QtCore import QThread, pyqtSignal
import numpy as np
import cv2

class DetectThread(QThread):
    callback = pyqtSignal(object)
    def __init__(self, model, file, parent=None):
        super().__init__(parent)
        self.model=model
        self.file=file

    def text(self, img, text, xy=(0, 0), color=(0, 0, 0), size=12):
        pil = Image.fromarray(img)
        s = platform.system()
        if s == "Linux":
            font = ImageFont.truetype(
                '/usr/share/fonts/truetype/wqy/wqy-zenhei.ttc',
                size)
        elif s == "Darwin":
            font = ImageFont.truetype('....', size)
        else:
            font = ImageFont.truetype('simsun.ttc', size)
        ImageDraw.Draw(pil).text(xy, text, font=font, fill=color)
        return np.array(pil)
    def run(self):
        img = cv2.imdecode(
            np.fromfile(self.file, dtype=np.uint8),
            cv2.IMREAD_COLOR)[:,:,::-1].copy()
        results = self.model.predict(img)[0]
        names = [results.names[int(i.cpu().numpy())] for i in results.boxes.cls]
        boxes = results.boxes.xyxy

        for box, name in zip(boxes, names):
            box = box.cpu().numpy()
            x1 = int(box[0])
            y1 = int(box[1])
            x2 = int(box[2])
            y2 = int(box[3])
            img = cv2.rectangle(img, (x1, y1), (x2, y2), (0, 255, 0), 3)
            img = self.text(img, name, (x1, y1 - 20), color=(0, 0, 255), size=100)
        self.callback.emit(img)

主程式

主程式代碼如下

import sys
import time
from PyQt6.QtCore import QSize
from PyQt6.QtGui import QIcon, QPixmap, QImage
from PyQt6.QtWidgets import QMainWindow, QApplication, QPushButton, QListWidgetItem, QFileDialog
from DetectThread import DetectThread
from ModelThread import ModelThread
from PictureThread import PictureThread
from ui.ui_mainwindow import Ui_MainWindow

class YoloV8App(QMainWindow, Ui_MainWindow):
    def __init__(self, parent=None):
        super().__init__(parent)
        self.setupUi(self)
        self.resize(1920,1080)
        self.btnPath.clicked.connect(self.btnPath_click)
        self.lblPath.setText("D:/pictures/primitive/2020/20200126_Philippines_day5_Boracay")
        self.path=self.lblPath.text()

        self.lblStatus.setText("載入模型中, 請稍後.....")
        self.modelThread=ModelThread()
        self.modelThread.callback.connect(self.ModelThreadCallback)
        self.modelThread.start()
    def btnPath_click(self):
        path=QFileDialog.getExistingDirectory()
        if path!='':
            self.path=path.replace("\\","/")
            self.lblPath.setText(self.path)
            self.listWidget.clear()
            self.pictureThread.runFlag=False
            self.pictureThread=PictureThread(self.path)
            self.pictureThread.callback.connect(self.pictureThreadCallback)
            self.pictureThread.start()
    def ModelThreadCallback(self, model):
        self.model=model
        self.lblStatus.setText("")
        self.pictureThread=PictureThread(self.path)
        self.pictureThread.callback.connect(self.pictureThreadCallback)
        self.pictureThread.start()
    def pictureThreadCallback(self, pix):
        btn=QPushButton()
        btn.setIcon(QIcon(pix))
        btn.setIconSize(QSize(400,300))
        btn.tag=pix.tag
        btn.clicked.connect(self.btn_click)
        item=QListWidgetItem()
        item.setSizeHint(QSize(400,300))
        self.listWidget.addItem(item)
        self.listWidget.setItemWidget(item, btn)
    def btn_click(self):
        btn=self.sender()
        self.lblStatus.setText('yolov8 辨識中.....')
        self.t1=time.time()
        self.detectThread=DetectThread(self.model, btn.tag)
        self.detectThread.callback.connect(self.detectThreadCallback)
        self.detectThread.start()
    def detectThreadCallback(self, img):
        self.t2=time.time()
        self.lblStatus.setText(f'辨識時間 {self.t2-self.t1:.7f}秒')
        pix=QPixmap(
            QImage(img,
                   img.shape[1],
                   img.shape[0],
                   img.shape[1]*3,
                   QImage.Format.Format_RGB888
                   )
            )
        pr=pix.width()/pix.height()
        lr=self.lblImg.width()/self.lblImg.height()
        if pr>lr:
            pix=pix.scaled(self.lblImg.width(), int(self.lblImg.width()/pr))
        else:
            pix=pix.scaled(int(self.lblImg.height()*lr), self.lblImg.height())
        self.lblImg.setPixmap(pix)
    def closeEvent(self, evnet):
        if self.pictureThread is not None:
            self.pictureThread.runFlag=False
app=QApplication(sys.argv)
mainWindow=YoloV8App()
mainWindow.showMaximized()
app.exec()