在PyQt中基于BB坐标实现bounding box点击选中以标注对象名称

核心修改点 #

1. 调整数据存储逻辑，将同一张图片的所有bounding box（以下简称bbox）按路径分组，解决单图多bbox的展示问题
2. 给显示图片的Label添加点击事件监听，实现点击坐标到图片实际像素坐标的转换
3. 实现bbox碰撞检测逻辑，判断点击坐标落在哪个bbox范围内
4. 新增选中状态高亮，选中的bbox用不同颜色绘制
5. 新增命名按钮绑定逻辑，支持给选中的bbox赋值名称

完整修改后代码 #

from PyQt5 import QtGui, QtWidgets, QtCore
from PyQt5.QtWidgets import QFileDialog, QInputDialog, QApplication
import csv
from pygui import Ui_MainWindow
from collections import defaultdict
import sys
import cv2

# 给Row新增可选的name字段
class Row:
    def __init__(self, image_path, x, y, w, h, name=None):
        self.image_path = image_path
        self.x = int(x)
        self.y = int(y)
        self.w = int(w)
        self.h = int(h)
        self.name = name

class mainProgram(QtWidgets.QMainWindow, Ui_MainWindow):
    def __init__(self, parent=None):
        super(mainProgram, self).__init__(parent)
        self.setupUi(self)
        # 初始化存储变量
        self.image_groups = defaultdict(list) # 按图片路径分组存储所有bbox
        self.image_list = [] # 所有不重复的图片路径列表，用于切换
        self.current_img_index = 0 # 当前显示的图片索引
        self.current_bboxes = [] # 当前图片的所有bbox
        self.selected_bbox = None # 当前选中的bbox
        # 给label安装事件过滤器，监听点击事件
        self.label.installEventFilter(self)

    def all_callbacks(self):
        self.Upload.clicked.connect(self.on_click_upload)
        self.Next.clicked.connect(self.on_click_next)
        self.Previous.clicked.connect(self.on_click_previous)
        # 新增命名按钮的回调，替换为你UI里命名按钮的实际对象名
        self.NameBtn.clicked.connect(self.on_click_name)

    def eventFilter(self, watched, event):
        # 监听label的鼠标点击事件
        if watched == self.label and event.type() == QtCore.QEvent.MouseButtonPress:
            if not self.label.pixmap() or len(self.current_bboxes) == 0:
                return super().eventFilter(watched, event)
            # 坐标转换：label点击坐标转图片实际像素坐标
            pixmap = self.label.pixmap()
            label_w = self.label.width()
            label_h = self.label.height()
            pixmap_w = pixmap.width()
            pixmap_h = pixmap.height()
            scale_x = pixmap_w / label_w
            scale_y = pixmap_h / label_h
            click_x = event.x() * scale_x
            click_y = event.y() * scale_y
            # 碰撞检测：判断点击落在哪个bbox里
            self.selected_bbox = None
            # 倒序遍历，优先选中最上层的bbox（如果有重叠）
            for bbox in reversed(self.current_bboxes):
                x1, y1 = bbox.x, bbox.y
                x2, y2 = bbox.x + bbox.w, bbox.y + bbox.h
                if x1 <= click_x <= x2 and y1 <= click_y <= y2:
                    self.selected_bbox = bbox
                    break
            # 重绘当前图片，高亮选中的bbox
            self.draw_current_image()
            return True
        return super().eventFilter(watched, event)

    def convert_cv_image_to_qt(self, cv_img):
        rgb_image = cv2.cvtColor(cv_img, cv2.COLOR_BGR2RGB)
        h, w, ch = rgb_image.shape
        bytes_per_line = ch * w
        convert_to_Qt_format = QtGui.QImage(rgb_image.data, w, h, bytes_per_line, QtGui.QImage.Format_RGB888)
        return QtGui.QPixmap.fromImage(convert_to_Qt_format)

    def draw_current_image(self):
        if len(self.current_bboxes) == 0:
            return
        image_path = self.current_bboxes[0].image_path
        image = cv2.imread(image_path)
        # 先画所有未选中的bbox，用红色
        for bbox in self.current_bboxes:
            if bbox != self.selected_bbox:
                cv2.rectangle(image, (bbox.x, bbox.y), (bbox.x+bbox.w, bbox.y+bbox.h), (0,0,255), 2)
                # 如果有名称，显示名称
                if bbox.name:
                    cv2.putText(image, bbox.name, (bbox.x, bbox.y-10), cv2.FONT_HERSHEY_SIMPLEX, 0.9, (0,0,255), 2)
        # 再画选中的bbox，用绿色，层级最高
        if self.selected_bbox:
            bbox = self.selected_bbox
            cv2.rectangle(image, (bbox.x, bbox.y), (bbox.x+bbox.w, bbox.y+bbox.h), (0,255,0), 2)
            text = bbox.name if bbox.name else "未命名"
            cv2.putText(image, text, (bbox.x, bbox.y-10), cv2.FONT_HERSHEY_SIMPLEX, 0.9, (0,255,0), 2)
        qimage = self.convert_cv_image_to_qt(image)
        self.label.setPixmap(qimage)
        self.label.show()

    def on_click_upload(self):
        dialog = QFileDialog()
        csv_file = dialog.getOpenFileName(None, "Import CSV", "", "CSV data files (*.csv)")
        if not csv_file[0]:
            return
        try:
            with open(csv_file[0], encoding='utf-8') as fp:
                reader = csv.reader(fp, delimiter=',')
                self.image_groups.clear()
                for r in reader:
                    if len(r) <5:
                        continue
                    row = Row(*r[:5])
                    self.image_groups[row.image_path].append(row)
        except PermissionError:
            print("你没有权限打开该文件")
            return
        if len(self.image_groups) ==0:
            print("文件为空，请选择其他文件")
            return
        self.image_list = list(self.image_groups.keys())
        self.current_img_index = 0
        self.current_bboxes = self.image_groups[self.image_list[self.current_img_index]]
        self.selected_bbox = None
        self.draw_current_image()

    def next_image(self, offset=1):
        if len(self.image_list) ==0:
            return
        self.current_img_index = (self.current_img_index + offset) % len(self.image_list)
        self.current_bboxes = self.image_groups[self.image_list[self.current_img_index]]
        self.selected_bbox = None
        self.draw_current_image()

    def on_click_next(self):
        self.next_image(offset=1)

    def on_click_previous(self):
        self.next_image(offset=-1)

    def on_click_name(self):
        # 点击命名按钮的逻辑
        if not self.selected_bbox:
            print("请先选中一个bounding box")
            return
        name, ok = QInputDialog.getText(self, "输入名称", "请输入目标对象名称：")
        if ok and name.strip():
            self.selected_bbox.name = name.strip()
            self.draw_current_image()

def execute_pipeline():
    app = QApplication(sys.argv)
    annotationGui = mainProgram()
    annotationGui.show()
    annotationGui.all_callbacks()
    sys.exit(app.exec_())

if __name__ == "__main__":
    execute_pipeline()

注意事项 #

• 代码中的NameBtn需要替换为你自己UI文件中命名按钮的实际对象名
• 导入的CSV文件每行格式仍为图片路径,x,y,w,h即可，名称会存在程序运行内存中，如需持久化存储可自行添加导出CSV的逻辑
• 若bbox存在重叠，点击重叠区域会优先选中后绘制的bbox（即列表中靠后的bbox）

内容的提问来源于stack exchange，提问作者iamkk