PaddleHub是飞浆预训练模型应用工具，它解决了人工智能应用中的数据量少引起的模型训练效果差的问题。目前PaddleHub模型库里面已有120多个模型，现有Python代码调用和命令行调用两种使用方法。本文主要用飞浆预训练模型实现图像分类、图像分割、人脸检测、目标检测，并进步一部实现PaddleHub模型在线部署。

环境准备

1.python3.6以上
2.安装padlepaddle1.7.1安装链接(https://paddlepaddle.org.cn/install/quick)
3.安装paddlehub$: pip install paddlehub -i https://mirrors.aliyun.com/pypi/simple/
4.查看模型库(https://paddlepaddle.org.cn/hublist)

图像分类-imagenet数据集

1、python代码调用#导入模块import paddlehub as hub#导入模型，第一次调用会自动下载module = hub.Module(name="xception71_imagenet")   test_img_path = "images/dog.jpg"#set input dict  input_dict = {"image": [test_img_path]}#execute predict and print the result  results = module.classification(data=input_dict) for result in results:print(result)2、命令调用：$:hub run xception71_imagenet --input_path test_detection.jpg

图像分割-人像分割

1、python代码调用import paddlehub as hubimport cv2humanseg = hub.Module(name="deeplabv3p_xception65_humanseg")   test_img_path = "images/1.jpg"#execute predict and print the result    results = humanseg.segmentation(paths=[test_img_path],visualization=True,output_dir="face_detection_result")
2、命令调用：$:hub run deeplabv3p_xception65_humanseg --input_path test_detection.jpg

目标检测-口罩检测

1、python代码调用import paddlehub as hubimport cv2module = hub.Module(name="pyramidbox_lite_server_mask")test_img_path = "./images/1.jpg"#set input dictinput_dict = {"data": [cv2.imread(test_img_path)]}results = module.face_detection(data=input_dict)print(results)for result in results:print(result['data'])2、命令调用：$:hub run pyramidbox_lite_mobile_mask --input_path test_detection.jpg

目标检测-人脸检测

1、python代码调用import paddlehub as hubmodule = hub.Module(name="pyramidbox_face_detection")test_img_path = "images/57.jpg"#execute predict and print the resultresults = module.face_detection(paths=[test_img_path],visualization=True,output_dir="face_detection_result")2、命令调用：$:hub run pyramidbox_face_detection --input_path test_detection.jpg

Paddlehub模型在线部署

1.启动PaddleHub Serving

  $ hub serving start -m pyramidbox_lite_server_mask

2.client端请求

 import requestsimport jsonimport cv2import base64import osfrom PIL import Image, ImageDrawdef get_save_image_name(output_dir, image_path):"""Get save image name from source image path."""if not os.path.exists(output_dir):os.makedirs(output_dir)image_name = os.path.split(image_path)[-1]name, ext = os.path.splitext(image_name)return os.path.join(output_dir, "{}".format(name)) + extdef draw_bounding_box_on_image(img_path,save_im_path, output_data):image = Image.open(img_path)draw = ImageDraw.Draw(image)for bbox in output_data:# draw bouding boxif bbox['label'] == "MASK":draw.line([(bbox['left'], bbox['top']),(bbox['left'], bbox['bottom']),(bbox['right'], bbox['bottom']),(bbox['right'], bbox['top']),(bbox['left'], bbox['top'])],width=2,fill='green')else:draw.line([(bbox['left'], bbox['top']),(bbox['left'], bbox['bottom']),(bbox['right'], bbox['bottom']),(bbox['right'], bbox['top']),(bbox['left'], bbox['top'])],width=2,fill='red')# draw labeltext = bbox['label'] + ": %.2f%%" % (100 * bbox['confidence'])textsize_width, textsize_height = draw.textsize(text=text)if image.mode == 'RGB' or image.mode == 'RGBA':box_fill = (255, 255, 255)text_fill = (0, 0, 0)else:box_fill = (255)text_fill = (0)draw.rectangle(xy=(bbox['left'], bbox['top'] - (textsize_height + 5),bbox['left'] + textsize_width + 10, bbox['top'] - 3),fill=box_fill)draw.text(xy=(bbox['left'], bbox['top'] - 15), text=text, fill=text_fill)return imagedef cv2_to_base64(image):data = cv2.imencode('.jpg', image)[1]return base64.b64encode(data.tostring()).decode('utf8')#发送HTTP请求img_path = "images/2.jpg"output_dir = "detection_result/"org_im = cv2.imread(img_path)data = {'images':[cv2_to_base64(cv2.imread("images/2.jpg"))]}headers = {"Content-type": "application/json"}url = "http://127.0.0.1:8866/predict/pyramidbox_lite_server_mask"r = requests.post(url=url, headers=headers, data=json.dumps(data))#打印预测结果print(r.json()["results"])#后期处理save_name = get_save_image_name(output_dir, img_path)
image = draw_bounding_box_on_image(img_path,output_dir, r.json()["results"][0]['data'])#保存结果image.save(save_name,quality=95)

参考文档

1、https://paddlepaddle.org.cn/install/quick
2、https://paddlepaddle.org.cn/hublist