0 Prefacio
Al entrenar el modelo de detección de imágenes de PaddleDetect, es necesario etiquetar manualmente el conjunto de datos. A continuación, tomaremos la detección de camiones como ejemplo y registraremos el proceso detallado de etiquetado con labelme, para no olvidarlo en el futuro.
Descargue los archivos utilizados en este artículo aquí: https://download.csdn.net/download/loutengyuan/87616492
1 labelme medio ambiente construcción
labelme es una herramienta de anotación de imágenes gráficas, que está escrita en Python y utiliza Qt para su interfaz gráfica. Para decirlo sin rodeos, tiene una interfaz, como un software, y se puede interactuar con él, pero se inicia desde la línea de comandos, lo cual es un poco más problemático que usar software. Su interfaz es la siguiente:
Nota: he realizado la sincronización de la interfaz aquí. Originalmente estaba en inglés, y los archivos de sincronización se proporcionarán a continuación.
1.1 instalación de la herramienta labelme
Primero abra la ventana de comando cmd y luego ingrese el siguiente comando para instalar:
# -i 是指定国内镜像源，提高安装速度
pip install labelme -i https://mirror.baidu.com/pypi/simple
Compruebe si está instalado con el siguiente comando:
pip list
1.2 etiqueta china
Reemplace el archivo chino labelme (archivo app.py en el directorio de descarga) con el siguiente directorio (esto se basa en su directorio de instalación real):
C:\Users\Administrator\AppData\Local\Programs\Python\Python310\Lib\site-packages\labelme
2 Proceso de operación de etiquetado
Abra el archivo descargado, puede ver los siguientes archivos:
- truck_detect: la imagen a marcar
- app.py: archivo chino labelme
- etiquetas.txt: lista de etiquetas
- Haga doble clic para abrir la herramienta de anotación.bat: inicie el script
2.1 Abra la herramienta de anotación
Originalmente, el software labelme se inició a través de la línea de comando. Aquí puse este comando en el doble clic para abrir el archivo labelme tool.bat, así que simplemente haga doble clic en este script para abrir la herramienta labelme. El contenido del script de inicio es el siguiente:
:: 进入当前目录
cd %~dp0
labelme truck_detect --labels labels.txt
pause
2.2 Activar la opción de guardado automático
- Seleccione Archivo -> Guardar automáticamente para guardar automáticamente la opción
Nueva dimensión
Seleccione la opción Dibujar rectángulo o Dibujar polígono
para realizar los siguientes pasos, instale el tipo de etiqueta para marcar la ubicación de cada área en la imagen y luego haga clic en Siguiente, el área dibujada se guardará automáticamente:
2.3 Editar/modificar anotación
Si la etiqueta se selecciona incorrectamente o es necesario ajustar el área, puede hacer clic en Editar para seleccionar el área que se editará para modificar, o hacer clic con el botón derecho para seleccionar Editar etiqueta para modificar.
2.4 Anotar archivos
Los archivos e imágenes marcados se guardan en la carpeta truck_detect:
el archivo de etiqueta está en formato json y el contenido es el siguiente:
{
"version": "5.1.1",
"flags": {
},
"shapes": [
{
"label": "整车侧面",
"points": [
[
96.73611111111113,
66.625
],
[
446.0416666666667,
290.23611111111114
]
],
"group_id": null,
"shape_type": "rectangle",
"flags": {
}
},
{
"label": "车头侧面",
"points": [
[
101.59722222222224,
70.79166666666666
],
[
299.5138888888889,
294.40277777777777
]
],
"group_id": null,
"shape_type": "rectangle",
"flags": {
}
},
{
"label": "车轮侧面",
"points": [
[
177.29166666666669,
188.84722222222223
],
[
431.45833333333337,
288.84722222222223
]
],
"group_id": null,
"shape_type": "rectangle",
"flags": {
}
},
{
"label": "车身侧面",
"points": [
[
290.48611111111114,
70.79166666666666
],
[
439.09722222222223,
279.125
]
],
"group_id": null,
"shape_type": "rectangle",
"flags": {
}
}
],
"imagePath": "81675.jpg_800x533.jpg",
"imageData": "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",
"imageHeight": 327,
"imageWidth": 490
}
conversión de 3 formatos
Los conjuntos de datos de detección de objetivos de uso común vienen en dos formatos, VOC y COCO.
3.1 Convertir a formato COCO
Si usa el formato COCO, se recomienda usar x2coco en PaddleDetection para convertir el archivo marcado en un conjunto de datos en formato COCO. El código de conversión es el siguiente:
python tools/x2coco.py \
--dataset_type labelme \
--json_input_dir ./labelme_annos/ \
--image_input_dir ./labelme_imgs/ \
--output_dir ./cocome/ \
--train_proportion 0.8 \
--val_proportion 0.2 \
--test_proportion 0.0
3.2 Convertir a formato VOC
Este proyecto usa el conjunto de datos en formato VOC, use el siguiente código para convertir el conjunto de datos en formato VOC2007:
import os
import numpy as np
import codecs
import json
from glob import glob
import cv2
import shutil
# 标签路径
labelme_path = "D:/DataProcess/vehicle/truck_detect/truck_detect/" # 原始labelme标注数据路径
saved_path = "D:/DataProcess/vehicle/truck_detect/truck_detect_voc/" # 保存路径
# 将中文标签名称转换成英文字母
label_translate_dict = {
"车头正面": "head_front",
"车头侧面": "head_side",
"车身侧面": "body_side",
"车轮侧面": "wheel_side",
"整车侧面": "vehicle_side",
"工程车": "vehicle_side" # engine_truck
}
label_txt = ["__ignore__", "_background_"]
# 创建要求文件夹
if not os.path.exists(saved_path + "Annotations"):
os.makedirs(saved_path + "Annotations")
if not os.path.exists(saved_path + "JPEGImages/"):
os.makedirs(saved_path + "JPEGImages/")
# 获取待处理文件
files = glob(labelme_path + "*.json")
files = [i.split("\\")[-1].split(".json")[0] for i in files]
total_len = len(files)
total_idx = 0
out_count = 0
skip_count = 0
err_count = 0
# 读取标注信息并写入 xml
for json_file_ in files:
total_idx += 1
if " " in json_file_:
continue
json_filename = labelme_path + json_file_ + ".json"
img_filename = labelme_path + json_file_ + ".jpg"
if not os.path.exists(json_filename) or not os.path.exists(img_filename):
continue
json_file = json.load(open(json_filename, "r", encoding="utf-8"))
if total_idx % 100 == 0:
print("进度：{}".format((total_idx / total_len) * 100))
imagePath = json_file["imagePath"]
if json_file_ not in imagePath:
err_count += 1
print("imagePath error fileName = {} imagePath = {}".format(json_file_, imagePath))
continue
height, width, channels = cv2.imread(img_filename).shape
success_count = 0
xml_filename = saved_path + "Annotations/" + json_file_ + ".xml"
with codecs.open(xml_filename, "w", "utf-8") as xml:
xml.write('<annotation>\n')
xml.write('\t<folder>' + 'UAV_data' + '</folder>\n')
xml.write('\t<filename>' + json_file_ + ".jpg" + '</filename>\n')
xml.write('\t<source>\n')
xml.write('\t\t<database>The UAV autolanding</database>\n')
xml.write('\t\t<annotation>UAV AutoLanding</annotation>\n')
xml.write('\t\t<image>flickr</image>\n')
xml.write('\t\t<flickrid>NULL</flickrid>\n')
xml.write('\t</source>\n')
xml.write('\t<owner>\n')
xml.write('\t\t<flickrid>NULL</flickrid>\n')
xml.write('\t\t<name>ChaojieZhu</name>\n')
xml.write('\t</owner>\n')
xml.write('\t<size>\n')
xml.write('\t\t<width>' + str(width) + '</width>\n')
xml.write('\t\t<height>' + str(height) + '</height>\n')
xml.write('\t\t<depth>' + str(channels) + '</depth>\n')
xml.write('\t</size>\n')
xml.write('\t\t<segmented>0</segmented>\n')
for multi in json_file["shapes"]:
points = np.array(multi["points"])
xmin = min(points[:, 0])
xmax = max(points[:, 0])
ymin = min(points[:, 1])
ymax = max(points[:, 1])
if "未知" in multi["label"]:
break
for key in label_translate_dict.keys():
if key in multi["label"]:
label = label_translate_dict[key]
break
if label is None:
continue
if label not in label_txt:
label_txt.append(label)
if xmax <= xmin:
continue
elif ymax <= ymin:
continue
elif (xmax - xmin) * (ymax - ymin) < 1000:
continue
else:
xml.write('\t<object>\n')
xml.write('\t\t<name>' + label + '</name>\n')
xml.write('\t\t<pose>Unspecified</pose>\n')
xml.write('\t\t<truncated>1</truncated>\n')
xml.write('\t\t<difficult>0</difficult>\n')
xml.write('\t\t<bndbox>\n')
xml.write('\t\t\t<xmin>' + str(xmin) + '</xmin>\n')
xml.write('\t\t\t<ymin>' + str(ymin) + '</ymin>\n')
xml.write('\t\t\t<xmax>' + str(xmax) + '</xmax>\n')
xml.write('\t\t\t<ymax>' + str(ymax) + '</ymax>\n')
xml.write('\t\t</bndbox>\n')
xml.write('\t</object>\n')
# print(total_idx, total_len, json_filename, xmin, ymin, xmax, ymax, label)
success_count += 1
xml.write('</annotation>')
if success_count > 0:
shutil.copy(img_filename, saved_path + "JPEGImages/")
out_count += 1
else:
skip_count += 1
if os.path.exists(xml_filename):
os.remove(xml_filename)
# 写标签文件
labels_txt_writer = open(saved_path + '/labels.txt', 'w')
for text in label_txt:
labels_txt_writer.write(text + "\n")
labels_txt_writer.close()
print("-------------------------------- Finished --------------------------------")
print("out_count = {}".format(out_count))
print("skip_count = {}".format(skip_count))
print("err_count = {}".format(err_count))
print("label_txt = {}".format(label_txt))
3.3 Área de etiqueta de salida
El siguiente script se usa para analizar el archivo json de la etiqueta y recortar o dibujar el resultado en la imagen por separado para ver el efecto de etiquetado:
import os
import numpy as np
import json
from glob import glob
import cv2
from PIL import Image, ImageDraw, ImageFont
labelme_path = "D:/DataProcess/vehicle/truck_detect/truck_detect/" # 原始labelme标注数据路径
# 输出文件夹路径
outputPath = "D:/DataProcess/vehicle/predict_output/"
files = glob(labelme_path + "*.json")
files = [i.split("\\")[-1].split(".json")[0] for i in files]
files_size = len(files)
# 将区域裁剪单独保存
def clip_save():
file_idx = 0
for json_file_ in files:
file_idx += 1
if " " in json_file_:
continue
json_filename = labelme_path + json_file_ + ".json"
img_filename = labelme_path + json_file_ + ".jpg"
if not os.path.exists(json_filename) or not os.path.exists(img_filename):
continue
json_file = json.load(open(json_filename, "r", encoding="utf-8"))
# image = cv2.imread(img_filename)
image = cv2.imdecode(np.fromfile(img_filename, dtype=np.uint8), -1)
# height, width, channels = image.shape
# "车头正面", "车头侧面", "车身侧面", "车轮侧面", "整车侧面", "工程车"
for label_name in ["车头正面", "车头侧面"]:
outputLabelPath = outputPath + label_name + "/"
if not os.path.exists(outputLabelPath):
os.makedirs(outputLabelPath)
idx = 0
for multi in json_file["shapes"]:
points = np.array(multi["points"])
xmin = min(points[:, 0])
xmax = max(points[:, 0])
ymin = min(points[:, 1])
ymax = max(points[:, 1])
# "车头正面", "车头侧面", "车身侧面", "车轮侧面", "整车侧面"
label = multi["label"]
if label_name not in label:
continue
if xmax <= xmin:
continue
elif ymax <= ymin:
continue
elif (xmax - xmin) * (ymax - ymin) < 1000:
continue
else:
print("{}/{} {} {} {} {} {} {}".format(file_idx, files_size, json_filename, xmin, ymin, xmax, ymax,
label))
# 将主体区域裁剪后保存到输出文件夹 #参数1 是高度的范围，参数2是宽度的范围
img_crop = image[int(ymin):int(ymax), int(xmin):int(xmax)]
if idx == 0:
out_image_file = outputLabelPath + json_file_ + ".jpg"
else:
out_image_file = outputLabelPath + "crop_" + str(idx) + "_" + json_file_ + ".jpg"
# cv2.imwrite(out_image_file, img_crop)
cv2.imencode('.jpg', img_crop)[1].tofile(out_image_file)
idx += 1
# 将结果绘制到一张图片
def draw_save():
file_idx = 0
for json_file_ in files:
file_idx += 1
if " " in json_file_:
continue
json_filename = labelme_path + json_file_ + ".json"
img_filename = labelme_path + json_file_ + ".jpg"
if not os.path.exists(json_filename) or not os.path.exists(img_filename):
continue
json_file = json.load(open(json_filename, "r", encoding="utf-8"))
# image = cv2.imread(img_filename)
image = cv2.imdecode(np.fromfile(img_filename, dtype=np.uint8), -1)[:, :, ::-1]
if isinstance(image, np.ndarray):
image = Image.fromarray(image)
# height, width, channels = image.shape
is_success = False
for multi in json_file["shapes"]:
points = np.array(multi["points"])
xmin = min(points[:, 0])
xmax = max(points[:, 0])
ymin = min(points[:, 1])
ymax = max(points[:, 1])
# "车头正面", "车头侧面", "车身侧面", "车轮侧面", "整车侧面"
label = multi["label"]
tag = False
for text in ["车头正面"]:
if text in label:
tag = True
break
if not tag:
continue
if xmax <= xmin:
continue
elif ymax <= ymin:
continue
else:
is_success = True
print("{}/{} {} {} {} {} {} {}".format(file_idx, files_size, json_filename, xmin, ymin, xmax, ymax,
label))
draw = ImageDraw.Draw(image)
font_size = 36
# simfang.ttf 这个文件在下载包中有
font = ImageFont.truetype("./simfang.ttf", font_size, encoding="utf-8")
text = "标签：{}".format(multi["label"])
th = font_size
tw = font.getsize(text)[0]
# tw = int(len(result["rec_docs"]) * font_size) + 60
start_y = max(0, ymin - th)
draw.rectangle(
[(xmin + 1, start_y), (xmin + tw + 1, start_y + th)], fill=(0, 102, 255))
draw.text((xmin + 1, start_y), text, fill=(255, 255, 255), font=font)
draw.rectangle(
[(xmin, ymin), (xmax, ymax)], outline=(255, 0, 0), width=2)
if is_success:
# image.show(img_filename)
os.makedirs(outputPath, exist_ok=True)
output_path = os.path.join(outputPath, json_file_ + ".jpg")
image.save(output_path, quality=95)
if __name__ == '__main__':
# 将结果单独裁剪保存
clip_save()
# # 将结果绘制到同一张图片上保存
# draw_save()