如何在4x gpu上运行opencv并进行多处理而不出错

Question

我在opencv和yolov3上遇到了一个小问题。我用CUDA编译了opencv。一切都很好，我在几个进程上运行推断，一切都在图形卡上运行。 我的电脑里有4张显卡。 如果我使用一个进程，那么选择脚本应该使用的图形卡是没有问题的

cv2.cuda.setDevice (0-3) 

很好。但在一个依赖于多个进程的脚本中，我加入了图形卡选择，得到了一个错误

(-217: Gpu API call) initialization error in function 'ManagedPtr'

import cv2
import numpy as np
from datetime import datetime
import os

class YoloEval(object):
  def __init__(self, requestedProbability):
    print("LOADING YOLO")
    cv2.cuda.setDevice(1)    
    self.net = cv2.dnn.readNet("yolov3.weights", "yolov3.cfg")
    self.net.setPreferableBackend(cv2.dnn.DNN_BACKEND_CUDA)
    self.net.setPreferableTarget(cv2.dnn.DNN_TARGET_CUDA_FP16)
    self.layer_names = self.net.getLayerNames()
    self.output_layers = [self.layer_names[i[0]-1] for i in self.net.getUnconnectedOutLayers()]
    self.requestedProbability = requestedProbability
    print("YOLO LOADED")
    self.saveImg = True

 def predict(self, imageToPredict, locationName, cameraId):
    height, width, channels = imageToPredict.shape
    blob = cv2.dnn.blobFromImage(imageToPredict, 1/255.0, (416,416), swapRB=True, crop=False)
    self.net.setInput(blob)
    outs = self.net.forward(self.output_layers)
    class_ids = []
    confidences = []
    boxes = []
    detected = False
    for out_index, out in enumerate(outs):
      for detection_index, detection in enumerate(out):
        scores = detection[5:]
        class_id = np.argmax(scores)
        confidence = scores[class_id]
        if confidence > self.requestedProbability and class_id == 0:
          center_x = int(detection[0] * width)
          center_y = int(detection[1] * height)
          w = int(detection[2] * width)
          h = int(detection[3] * height)
          x = int(center_x - w / 2)
          y = int(center_y - h /2)
          boxes.append([x, y, w, h])
          confidences.append(float(confidence))
          class_ids.append(class_id)
          indexes = cv2.dnn.NMSBoxes(boxes, confidences, self.requestedProbability, 0.4)
          font = cv2.FONT_HERSHEY_PLAIN
          colors = np.random.uniform(0, 255, size=(1, 3))
          for i in range(len(boxes)):
            if i in indexes:
              x, y, w, h = boxes[i]
              label = "Osoba"
              color = colors[class_ids[i]]
              cv2.rectangle(imageToPredict, (x, y), (x + w, y + h), color, 2)
              cv2.putText(imageToPredict, label, (x, y + 30), font, 2, color, 3)
          detected = True
    if detected: 
      return 1
    else: return 0