转自：Flask部署YOLOv5 - 知乎

YOLOv5的flask部署 - 迷途小书童的Note的个人空间 - OSCHINA - 中文开源技术交流社区

Flask是一种用python实现轻量级的web服务，也称为微服务，其灵活性较强而且效率高，在深度学习方面，也常常用来部署B/S模型。下面以yolov5s模型为例，介绍基于Flask的封装和部署过程。

• 封装YOLOv5

编写yolov5.py，封装yolov5推理过程

class YOLOv5(object):
    # 参数设置
    _defaults = {
        "weights": "./weights/yolov5s.pt",
        "imgsz": 640,
        "iou_thres":0.45,
        "conf_thres":0.25,
        "classes":0   #只检测人
    }

    @classmethod
    def get_defaults(cls,n):
        if n in cls._defaults:
            return cls._defaults[n]
        else:
            return "Unrecognized attribute name '" + n + "'"
    # 初始化操作，加载模型
    def __init__(self,device='0',**kwargs):
        self.__dict__.update(self._defaults)
        self.device = select_device(device)
        self.half = self.device != "cpu" 

        self.model = attempt_load(self.weights, map_location=self.device)  # load FP32 model
        self.imgsz = check_img_size(self.imgsz, s=self.model.stride.max())  # check img_size
        if self.half:
            self.model.half()  # to FP16
    
    # 推理部分
    def infer(self,inImg):
        # 使用letterbox方法将图像大小调整为640大小
        img = letterbox(inImg, new_shape=self.imgsz)[0]

        # 归一化与张量转换
        img = img[:, :, ::-1].transpose(2, 0, 1)  # BGR to RGB, to 3x416x416
        img = np.ascontiguousarray(img)
        img = torch.from_numpy(img).to(self.device)
        img = img.half() if self.half else img.float()  # uint8 to fp16/32
        img /= 255.0  # 0 - 255 to 0.0 - 1.0
        if img.ndimension() == 3:
            img = img.unsqueeze(0)

        # 推理
        pred = self.model(img, augment=True)[0]
        # NMS
        pred = non_max_suppression(pred, self.conf_thres, self.iou_thres, classes=self.classes, agnostic=True)
        
        bbox_xyxy = []
        confs = []
        cls_ids = []

        # 解析检测结果
        for i, det in enumerate(pred):  # detections per image
            if det is not None and len(det):
                # 将检测框映射到原始图像大小
                det[:, :4] = scale_coords(img.shape[2:], det[:, :4], inImg.shape).round()
                # 保存结果
                for *xyxy, conf, cls in reversed(det):
                    bbox_xyxy.append(xyxy)
                    confs.append(conf.item())
                    cls_ids.append(int(cls.item()))
    
                xyxys = torch.Tensor(bbox_xyxy)
                confss = torch.Tensor(confs)
                cls_ids = torch.Tensor(cls_ids)

        return xyxys, confss, cls_ids

• 服务端程序

编写server.py文件，封装服务端程序

app = Flask(__name__)
det = YOLOv5()

@app.route("/infer", methods=["POST"])
def predict():
    result = {"success": False}
    if request.method == "POST":
        if request.files.get("image") is not None:
            try:
                # 得到客户端传输的图像          
                start = time.time()      
                input_image = request.files["image"].read()
                imBytes = np.frombuffer(input_image, np.uint8)
                iImage = cv2.imdecode(imBytes, cv2.IMREAD_COLOR)
                # 执行推理
                outs = det.infer(iImage)
                print("duration: ",time.time()-start)
 
                if (outs is None) and (len(outs) < 0):
                    result["success"] = False
                # 将结果保存为json格式
                result["box"] = outs[0].tolist()
                result["conf"] = outs[1].tolist()
                result["classid"] = outs[2].tolist()
                result['success'] = True
            
            except Exception:
                pass
        
    return jsonify(result)

if __name__ == "__main__":
    print(("* Loading yolov5 model and Flask starting server..."
        "please wait until server has fully started"))
    app.run(host='127.0.0.1', port=7000)

• 客户端程序

编写client.py，封装客户端程序

# 将图像以jpg编码，并转换为字节流
def get_img_bytes(img):  
    img_str = cv2.imencode('.jpg',img)[1].tobytes() if img is not None else None
    return img_str

# 定义工具方法，在原始图像上画框
def plot_one_box(x, img, color=None, label="person", line_thickness=None):
    """ 画框,引自 YoLov5 工程.
    参数: 
        x:      框， [x1,y1,x2,y2]
        img:    opencv图像
        color:  设置矩形框的颜色, 比如 (0,255,0)
        label:  str
        line_thickness: int
    return:
        no return
    """
    tl = (
        line_thickness or round(0.002 * (img.shape[0] + img.shape[1]) / 2) + 1
    )  # line/font thickness
    color = color or [random.randint(0, 255) for _ in range(3)]
    c1, c2 = (int(x[0]), int(x[1])), (int(x[2]), int(x[3]))
    cv2.rectangle(img, c1, c2, color, thickness=tl, lineType=cv2.LINE_AA)
    if label:
        tf = max(tl - 1, 1)  # font thickness
        t_size = cv2.getTextSize(label, 0, fontScale=tl / 3, thickness=tf)[0]
        c2 = c1[0] + t_size[0], c1[1] - t_size[1] - 3
        cv2.rectangle(img, c1, c2, color, -1, cv2.LINE_AA)  # filled
        cv2.putText(
            img,
            label,
            (c1[0], c1[1] - 2),
            0,
            tl / 3,
            [225, 255, 255],
            thickness=tf,
            lineType=cv2.LINE_AA,
        )
def main():  
     img = cv2.imread("./bus.jpg")
     bFrame = get_img_bytes(img)
     request_input = {'image': bFrame}
     result = requests.post('http://127.0.0.1:7000/infer', files=request_input).json()
     if result['success']:
         boxs = result["box"] 
         confs = result["conf"]
         ids = result["classid"] 
         
     if boxs is not None:
            for i,box in enumerate(boxs):
                plot_one_box(toInt(box),img,label=str(ids[i])
            cv2.imshow("image",img)
            cv2.waitKey(0)

• 效果展示