偷懒工具1：拒绝手动计算！我用 PySide6 撸了一个 OpenCV ROI 坐标提取“神器”
·
在搞工业视觉或者 OpenCV 开发时，最痛苦的时刻莫过于：为了提取一个 ROI 区域，得先用截图工具量像素，或者在代码里反复修改参数、运行、查看结果。
为了彻底告别“肉眼测距”，我基于 PySide6 开发了一个轻量化的图像标记工具。它能自动处理图像缩放、保持原始像素坐标，并支持矩形和多边形标注。最重要的是，拿到的坐标可以直接丢给 AI（如 Gemini 或 ChatGPT），让它一键生成 OpenCV 裁剪代码，生产力瞬间拉满！
结语
这个小工具虽然简单，但它解决了 CV 工程师每天都要面对的“微小琐事”。如果你也厌倦了手动去 OpenCV 里试坐标，不妨试试这个思路。
代码：
import sys
import os
from PySide6.QtWidgets import (
QApplication, QMainWindow, QWidget, QVBoxLayout, QHBoxLayout,
QPushButton, QFileDialog, QLabel, QScrollArea, QMessageBox, QSizePolicy,
QTextBrowser
)
from PySide6.QtGui import QPixmap, QPainter, QPen, QColor, QBrush, QPolygon
from PySide6.QtCore import Qt, QPoint, QRectF, QEvent
class ImageWidget(QLabel):
def __init__(self, parent=None):
super().__init__(parent)
self.setBackgroundRole(self.parent().backgroundRole())
self.setSizePolicy(QSizePolicy.Ignored, QSizePolicy.Ignored)
self.setScaledContents(False)
self.image_pixmap = QPixmap()
self.original_size = None
self.scale_factor = 1.0
self.offset_x = 0
self.offset_y = 0
self.current_tool = "None"
self.annotations = []
self.current_polygon = []
self.temp_rect_start = None
self.temp_rect_end = None
self.setMouseTracking(True)
def set_image(self, pixmap):
self.image_pixmap = pixmap
self.original_size = pixmap.size()
self.setFixedSize(self.original_size)
self.adjust_scale_parameters()
self.update()
def set_tool(self, tool_name):
if self.current_tool == "Polygon" and self.current_polygon:
self.complete_polygon()
self.current_tool = tool_name
self.current_polygon = []
self.temp_rect_start = None
self.temp_rect_end = None
self.update()
def clear_annotations(self):
self.annotations = []
self.current_polygon = []
self.temp_rect_start = None
self.temp_rect_end = None
self.update()
def get_original_coords(self, x_screen, y_screen):
if not self.original_size:
return None
x_mapped = x_screen - self.offset_x
y_mapped = y_screen - self.offset_y
x_orig = x_mapped / self.scale_factor
y_orig = y_mapped / self.scale_factor
x_orig = max(0, min(int(x_orig), self.original_size.width()))
y_orig = max(0, min(int(y_orig), self.original_size.height()))
return x_orig, y_orig
def get_screen_point(self, x_orig, y_orig):
x_screen = int(x_orig * self.scale_factor + self.offset_x)
y_screen = int(y_orig * self.scale_factor + self.offset_y)
return QPoint(x_screen, y_screen)
def adjust_scale_parameters(self):
if self.image_pixmap.isNull():
return
w_orig = self.original_size.width()
h_orig = self.original_size.height()
w_widget = self.width()
h_widget = self.height()
scale_w = w_widget / w_orig if w_orig > 0 else 1.0
scale_h = h_widget / h_orig if h_orig > 0 else 1.0
self.scale_factor = min(scale_w, scale_h)
displayed_w = w_orig * self.scale_factor
displayed_h = h_orig * self.scale_factor
self.offset_x = (w_widget - displayed_w) / 2
self.offset_y = (h_widget - displayed_h) / 2
def resizeEvent(self, event):
super().resizeEvent(event)
self.adjust_scale_parameters()
self.update()
def mousePressEvent(self, event):
if self.image_pixmap.isNull() or event.button() != Qt.LeftButton:
return
if event.type() == QEvent.Type.MouseButtonDblClick and len(self.current_polygon) >= 1:
self.complete_polygon()
return
point = event.position().toPoint()
x_orig, y_orig = self.get_original_coords(point.x(), point.y())
if x_orig is None: return
if self.current_tool == "Rect":
self.temp_rect_start = point
self.temp_rect_end = point
elif self.current_tool == "Polygon":
self.current_polygon.append((x_orig, y_orig))
self.update()
def mouseMoveEvent(self, event):
if self.image_pixmap.isNull():
return
if self.current_tool == "Rect" and self.temp_rect_start:
self.temp_rect_end = event.position().toPoint()
self.update()
elif self.current_tool == "Polygon" and self.current_polygon:
self.update()
def mouseReleaseEvent(self, event):
if self.image_pixmap.isNull() or event.button() != Qt.LeftButton:
return
if self.current_tool == "Rect" and self.temp_rect_start:
if self.temp_rect_start == self.temp_rect_end:
self.temp_rect_start = None
self.temp_rect_end = None
return
x1_orig, y1_orig = self.get_original_coords(self.temp_rect_start.x(), self.temp_rect_start.y())
x2_orig, y2_orig = self.get_original_coords(self.temp_rect_end.x(), self.temp_rect_end.y())
coords = [min(x1_orig, x2_orig), min(y1_orig, y2_orig),
max(x1_orig, x2_orig), max(y1_orig, y2_orig)]
self.annotations.append({'type': 'Rect', 'coords': coords})
self.temp_rect_start = None
self.temp_rect_end = None
self.update()
def complete_polygon(self):
if len(self.current_polygon) >= 3:
self.annotations.append({'type': 'Polygon', 'coords': self.current_polygon})
self.current_polygon = []
self.update()
self.current_tool = "None"
self.window().update_tool_buttons()
def paintEvent(self, event):
if self.image_pixmap.isNull():
return
painter = QPainter(self)
painter.setRenderHint(QPainter.Antialiasing)
target_rect = QRectF(self.offset_x, self.offset_y,
self.original_size.width() * self.scale_factor,
self.original_size.height() * self.scale_factor)
painter.drawPixmap(target_rect.toRect(), self.image_pixmap, self.image_pixmap.rect())
for ann in self.annotations:
if ann['type'] == 'Rect':
x1, y1, x2, y2 = ann['coords']
p1 = self.get_screen_point(x1, y1)
p2 = self.get_screen_point(x2, y2)
painter.setPen(QPen(QColor(255, 100, 0), 3))
painter.setBrush(Qt.NoBrush)
painter.drawRect(p1.x(), p1.y(), p2.x() - p1.x(), p2.y() - p1.y())
elif ann['type'] == 'Polygon':
poly = QPolygon()
for x_orig, y_orig in ann['coords']:
poly.append(self.get_screen_point(x_orig, y_orig))
painter.setPen(QPen(QColor(0, 255, 255), 3))
painter.setBrush(QBrush(QColor(0, 255, 255, 50)))
painter.drawPolygon(poly)
painter.setPen(QPen(QColor(255, 255, 0), 2, Qt.DashLine))
if self.current_tool == "Rect" and self.temp_rect_start:
painter.setBrush(Qt.NoBrush)
painter.drawRect(self.temp_rect_start.x(), self.temp_rect_start.y(),
self.temp_rect_end.x() - self.temp_rect_start.x(),
self.temp_rect_end.y() - self.temp_rect_start.y())
elif self.current_tool == "Polygon" and len(self.current_polygon) > 0:
poly = QPolygon()
for x_orig, y_orig in self.current_polygon:
p = self.get_screen_point(x_orig, y_orig)
poly.append(p)
painter.setBrush(QColor(255, 0, 0))
painter.drawEllipse(p, 5, 5)
if len(poly) > 1:
painter.setBrush(Qt.NoBrush)
painter.drawPolyline(poly)
painter.drawLine(poly.last(), self.mapFromGlobal(self.cursor().pos()))
painter.end()
def _create_description_box(self):
description_text = """
<h3>🎨 图像标记工具说明</h3>
<p>本工具用于在加载的图像上进行精确标记，并输出标记点的原始像素坐标。</p>
<h4>✅ 功能特点:</h4>
<ul>
<li><strong>自适应缩放:</strong> 图像等比例适应窗口大小。</li>
<li><strong>原始坐标输出:</strong> 标记点坐标不受缩放影响，始终输出原始像素值。</li>
<li><strong>多工具支持:</strong> 矩形框和多边形轮廓。</li>
</ul>
<h4>🔨 使用方法:</h4>
<h5>1. 矩形框标记</h5>
<p>点击 <strong>矩形框标记</strong> 按钮，然后在图像上：</p>
<ol>
<li>按住鼠标左键拖动。</li>
<li>释放鼠标，完成<strong>橙色</strong>矩形标记。</li>
</ol>
<h5>2. 多边形轮廓标记</h5>
<p>点击 <strong>多边形轮廓标记</strong> 按钮，然后在图像上：</p>
<ol>
<li>连续点击鼠标左键来设置轮廓点 (红点)。</li>
<li>完成绘制后，点击 <strong>完成多边形</strong> 按钮，或在图像上双击鼠标左键。</li>
</ol>
<h5>3. 输出与清空</h5>
<ul>
<li><strong>清空标记:</strong> 移除当前所有标记。</li>
<li><strong>输出并导出坐标:</strong> 将所有已完成标记的原始像素坐标保存到 <code>coordinates.txt</code>。</li>
</ul>
"""
text_browser = QTextBrowser()
text_browser.setHtml(description_text)
text_browser.setMaximumWidth(350)
text_browser.setMinimumWidth(250)
return text_browser
class MainWindow(QMainWindow):
def __init__(self):
super().__init__()
self.setWindowTitle("图像标记与坐标输出工具 (按 Esc 退出)") # 更新标题提示
self.setGeometry(100, 100, 1200, 700)
self.central_widget = QWidget()
self.setCentralWidget(self.central_widget)
self.main_layout = QVBoxLayout(self.central_widget)
# 1. 图像和描述区域 (水平布局)
self.content_layout = QHBoxLayout()
self.image_widget = ImageWidget(self)
self.scroll_area = QScrollArea()
self.scroll_area.setWidgetResizable(True)
self.scroll_area.setWidget(self.image_widget)
self.description_widget = self.image_widget._create_description_box()
self.content_layout.addWidget(self.scroll_area, 3) # 图像占 3/4 空间
self.content_layout.addWidget(self.description_widget, 1) # 描述框占 1/4 空间
self.main_layout.addLayout(self.content_layout)
# 2. 控制按钮区域
self.control_layout = QHBoxLayout()
self.main_layout.addLayout(self.control_layout)
self.setup_menu()
self.setup_buttons()
self.update_tool_buttons()
# ================= 修改部分 1：添加键盘事件监听 =================
def keyPressEvent(self, event):
"""监听键盘按键，如果按下 Esc，则关闭窗口"""
if event.key() == Qt.Key_Escape:
self.close()
else:
super().keyPressEvent(event)
# =============================================================
def setup_menu(self):
menu_bar = self.menuBar()
file_menu = menu_bar.addMenu("文件")
open_action = file_menu.addAction("打开图像")
open_action.triggered.connect(self.open_image)
def setup_buttons(self):
self.btn_rect = QPushButton("矩形框标记")
self.btn_rect.clicked.connect(lambda: self.set_tool("Rect"))
self.control_layout.addWidget(self.btn_rect)
self.btn_polygon = QPushButton("多边形轮廓标记 (点击完成)")
self.btn_polygon.clicked.connect(lambda: self.set_tool("Polygon"))
self.control_layout.addWidget(self.btn_polygon)
self.btn_complete_polygon = QPushButton("完成多边形 (双击鼠标也行)")
self.btn_complete_polygon.clicked.connect(self.image_widget.complete_polygon)
self.control_layout.addWidget(self.btn_complete_polygon)
self.control_layout.addStretch(1)
self.btn_clear = QPushButton("清空标记")
self.btn_clear.clicked.connect(self.image_widget.clear_annotations)
self.control_layout.addWidget(self.btn_clear)
self.btn_export = QPushButton("输出并导出坐标")
self.btn_export.clicked.connect(self.export_coordinates)
self.control_layout.addWidget(self.btn_export)
def update_tool_buttons(self):
tool = self.image_widget.current_tool
self.btn_rect.setStyleSheet("background-color: lightgreen;" if tool == "Rect" else "")
self.btn_polygon.setStyleSheet("background-color: lightgreen;" if tool == "Polygon" else "")
self.btn_complete_polygon.setEnabled(tool == "Polygon" or len(self.image_widget.current_polygon) >= 2)
def set_tool(self, tool_name):
self.image_widget.set_tool(tool_name)
self.update_tool_buttons()
def open_image(self):
file_name, _ = QFileDialog.getOpenFileName(self, "打开图像文件", "",
"Image Files (*.png *.jpg *.jpeg);;All Files (*)")
if file_name:
pixmap = QPixmap(file_name)
if pixmap.isNull():
QMessageBox.critical(self, "错误", f"无法加载图像文件，请检查文件路径和格式:\n{file_name}")
return
self.image_widget.set_image(pixmap)
self.image_widget.setFixedSize(self.image_widget.original_size)
self.image_widget.adjust_scale_parameters()
self.image_widget.clear_annotations()
self.set_tool("None")
self.setWindowTitle(f"图像标记与坐标输出工具 - {os.path.basename(file_name)} (按 Esc 退出)")
def export_coordinates(self):
annotations = self.image_widget.annotations
if not annotations:
QMessageBox.information(self, "提示", "没有检测到任何标记，请先进行标记。")
return
output = ["--- 图像标记原始像素坐标 (W={}, H={}) ---".format(
self.image_widget.original_size.width(), self.image_widget.original_size.height()
)]
for i, ann in enumerate(annotations):
ann_type = ann['type']
coords = ann['coords']
output.append(f"\n标记 {i + 1} ({ann_type}):")
if ann_type == 'Rect':
x_min, y_min, x_max, y_max = [int(c) for c in coords]
output.append(f" 类型: 矩形框")
output.append(f" 坐标: [{x_min}, {y_min}, {x_max}, {y_max}]")
output.append(f" 宽度: {x_max - x_min}, 高度: {y_max - y_min}")
elif ann_type == 'Polygon':
output.append(f" 类型: 多边形/轮廓 (共 {len(coords)} 个点)")
output.append(" 点列表 (X, Y):")
for j, (x, y) in enumerate(coords):
output.append(f" P{j + 1}: ({int(x)}, {int(y)})")
output_text = "\n".join(output)
print("\n" + "=" * 50)
print("--- 坐标输出开始 ---")
print(output_text)
print("--- 坐标输出结束 ---")
print("=" * 50 + "\n")
output_file = "coordinates.txt"
try:
with open(output_file, "w", encoding="utf-8") as f:
f.write(output_text)
QMessageBox.information(self, "导出成功", f"坐标已成功导出到:\n{os.path.abspath(output_file)}")
except Exception as e:
QMessageBox.critical(self, "导出错误", f"导出文件失败: {e}")
if __name__ == '__main__':
app = QApplication(sys.argv)
window = MainWindow()
window.show()
sys.exit(app.exec())
备注：只要用的话，我有封装好的通过网盘分享的文件：ROI原图标记工具.zip
链接: https://pan.baidu.com/s/1pQcRiRstkLDKImDTfo1L4Q?pwd=3h2e 提取码: 3h2e
--来自百度网盘超级会员v5的分享
AtomGit 是由开放原子开源基金会联合 CSDN 等生态伙伴共同推出的新一代开源与人工智能协作平台。平台坚持“开放、中立、公益”的理念，把代码托管、模型共享、数据集托管、智能体开发体验和算力服务整合在一起，为开发者提供从开发、训练到部署的一站式体验。
更多推荐
17
0- 0
已为社区贡献1条内容
所有评论(0)