基于多进程的TensorFlow目标检测机器人

本站内容均来自兴趣收集,如不慎侵害的您的相关权益,请留言告知,我们将尽快删除.谢谢.

欢迎大家学习我们的TensorFlow对象检测API教程系列的第10部分。首先，您可以在我的GitHub页面上下载代码。这将是这个CSGO aimbot视频系列的最后一个教程，因为现在，我在这个教程上花了太多的时间。尽管如此，我还是设法取得了目前最好的表现。为了进一步改进，我需要研究一下将来将使用的另一种检测方法–“EURO”。GitHub

在继续本教程之前，我应该提到，在将第9教程代码与CSGO TensorFlow检测代码合并之前，我更新了它。与多处理队列相比，性能是相同的(33FPS)，但是我想测试不同的数据方法来在进程之间共享数据。所以我更新了第9个教程，增加了一个文件，我们在其中使用多处理管道抓取屏幕。除了多处理管道之外，它们还使用一对一通信，队列可以用作“œ多对多”�。

继续本教程，我不会在此文本教程部分中进行代码解释。相反，我将整个代码分为3个部分：抓取屏幕、进行TensorFlow检测和显示屏幕。然后，这3个部分全部移到多处理进程中。

以下是最终代码：

# # Imports
import multiprocessing
from multiprocessing import Pipe
import time
import cv2
import mss
import numpy as np
import os
import sys
os.environ['CUDA_VISIBLE_DEVICES'] = '0'
import tensorflow as tf
from distutils.version import StrictVersion
from collections import defaultdict
from io import StringIO
import pyautogui
# title of our window
title = "FPS benchmark"
# set start time to current time
start_time = time.time()
# displays the frame rate every 2 second
display_time = 2
# Set primarry FPS to 0
fps = 0
# Load mss library as sct
sct = mss.mss()
# Set monitor size to capture to MSS
width = 800
height = 640
monitor = {"top": 80, "left": 0, "width": width, "height": height}
# ## Env setup
from object_detection.utils import ops as utils_ops
from object_detection.utils import label_map_util
from object_detection.utils import visualization_utils as vis_util
# # Model preparation 
PATH_TO_FROZEN_GRAPH = 'CSGO_frozen_inference_graph.pb'
# List of the strings that is used to add correct label for each box.
PATH_TO_LABELS = 'CSGO_labelmap.pbtxt'
NUM_CLASSES = 4
# ## Load a (frozen) Tensorflow model into memory.
label_map = label_map_util.load_labelmap(PATH_TO_LABELS)
categories = label_map_util.convert_label_map_to_categories(label_map, max_num_classes=NUM_CLASSES, use_display_name=True)
category_index = label_map_util.create_category_index(categories)
detection_graph = tf.Graph()
with detection_graph.as_default():
  od_graph_def = tf.GraphDef()
  with tf.gfile.GFile(PATH_TO_FROZEN_GRAPH, 'rb') as fid:
    serialized_graph = fid.read()
    od_graph_def.ParseFromString(serialized_graph)
    tf.import_graph_def(od_graph_def, name='')
def Shoot(mid_x, mid_y):
  x = int(mid_x*width)
  #y = int(mid_y*height)
  y = int(mid_y*height+height/9)
  pyautogui.moveTo(x,y)
  pyautogui.click()
def grab_screen(p_input):
  while True:
    #Grab screen image
    img = np.array(sct.grab(monitor))
    img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
    img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
    # Put image from pipe
    p_input.send(img)
def TensorflowDetection(p_output, p_input2):
  # Detection
  with detection_graph.as_default():
    with tf.Session(graph=detection_graph) as sess:
      while True:
        # Get image from pipe
        image_np = p_output.recv()
        # Expand dimensions since the model expects images to have shape: [1, None, None, 3]
        image_np_expanded = np.expand_dims(image_np, axis=0)
        # Actual detection.
        image_tensor = detection_graph.get_tensor_by_name('image_tensor:0')
        boxes = detection_graph.get_tensor_by_name('detection_boxes:0')
        scores = detection_graph.get_tensor_by_name('detection_scores:0')
        classes = detection_graph.get_tensor_by_name('detection_classes:0')
        num_detections = detection_graph.get_tensor_by_name('num_detections:0')
        # Visualization of the results of a detection.
        (boxes, scores, classes, num_detections) = sess.run(
            [boxes, scores, classes, num_detections],
            feed_dict={image_tensor: image_np_expanded})
        vis_util.visualize_boxes_and_labels_on_image_array(
            image_np,
            np.squeeze(boxes),
            np.squeeze(classes).astype(np.int32),
            np.squeeze(scores),
            category_index,
            use_normalized_coordinates=True,
            line_thickness=2)
        # Send detection image to pipe2
        p_input2.send(image_np)
        array_ch = []
        array_c = []
        array_th = []
        array_t = []
        for i,b in enumerate(boxes[0]):
          if classes[0][i] == 2: # ch
            if scores[0][i] >= 0.5:
              mid_x = (boxes[0][i][1]+boxes[0][i][3])/2
              mid_y = (boxes[0][i][0]+boxes[0][i][2])/2
              array_ch.append([mid_x, mid_y])
              cv2.circle(image_np,(int(mid_x*width),int(mid_y*height)), 3, (0,0,255), -1)
          if classes[0][i] == 1: # c 
            if scores[0][i] >= 0.5:
              mid_x = (boxes[0][i][1]+boxes[0][i][3])/2
              mid_y = boxes[0][i][0] + (boxes[0][i][2]-boxes[0][i][0])/6
              array_c.append([mid_x, mid_y])
              cv2.circle(image_np,(int(mid_x*width),int(mid_y*height)), 3, (50,150,255), -1)
          if classes[0][i] == 4: # th
            if scores[0][i] >= 0.5:
              mid_x = (boxes[0][i][1]+boxes[0][i][3])/2
              mid_y = (boxes[0][i][0]+boxes[0][i][2])/2
              array_th.append([mid_x, mid_y])
              cv2.circle(image_np,(int(mid_x*width),int(mid_y*height)), 3, (0,0,255), -1)
          if classes[0][i] == 3: # t
            if scores[0][i] >= 0.5:
              mid_x = (boxes[0][i][1]+boxes[0][i][3])/2
              mid_y = boxes[0][i][0] + (boxes[0][i][2]-boxes[0][i][0])/6
              array_t.append([mid_x, mid_y])
              cv2.circle(image_np,(int(mid_x*width),int(mid_y*height)), 3, (50,150,255), -1)
        team = "c" # shooting target
        if team == "c":
          if len(array_ch) > 0:
            Shoot(array_ch[0][0], array_ch[0][1])
          if len(array_ch) == 0 and len(array_c) > 0:
            Shoot(array_c[0][0], array_c[0][1])
        if team == "t":
          if len(array_th) > 0:
            Shoot(array_th[0][0], array_th[0][1])
          if len(array_th) == 0 and len(array_t) > 0:
            Shoot(array_t[0][0], array_t[0][1])
def Show_image(p_output2):
  global start_time, fps
  while True:
    image_np = p_output2.recv()
    # Show image with detection
    cv2.imshow(title, image_np)
    # Bellow we calculate our FPS
    fps+=1
    TIME = time.time() - start_time
    if (TIME) >= display_time :
      print("FPS: ", fps / (TIME))
      fps = 0
      start_time = time.time()
    # Press "q" to quit
    if cv2.waitKey(25) & 0xFF == ord("q"):
      cv2.destroyAllWindows()
      break
if __name__=="__main__":
    # Pipes
    p_output, p_input = Pipe()
    p_output2, p_input2 = Pipe()
    # creating new processes
    p1 = multiprocessing.Process(target=grab_screen, args=(p_input,))
    p2 = multiprocessing.Process(target=TensorflowDetection, args=(p_output,p_input2,))
    p3 = multiprocessing.Process(target=Show_image, args=(p_output2,))
    # starting our processes
    p1.start()
    p2.start()
    p3.start()

作为最终的结果，我很高兴我们可以达到20多个FPS。但是，当TensorFlow接收到我们探测到敌人的图像时，瓶颈就来了。FPS降到4？EURO“5帧/秒，我们的机器人玩这个游戏变得不可能了。因此，在未来，当我找到更快地发现我们的敌人的方法时，我可能会回到这个项目上来。有一种方法可以使用YOLO对象检测模型，它相当快速和准确，但它(目前)很难实现™。

不管怎幺说，我想我在这个项目上花了很多时间。现在，我将转到另一个更有益的项目。在不久的将来，我计划制作教程，介绍如何破解验证码™，如何使用硒来制作网络冲浪机器人或人工智能外汇交易机器人。

最初发表于https://pylessons.com/Tensorflow-object-detection-aim-bot-with-multiprocessing
https://pylessons.com/Tensorflow-object-detection-aim-bot-with-multiprocessing