How do you speed up the detection of the object or code as a whole?

[simrah]

I tried to do something like a yolov-based auto-driver. And it's kind of working, he's more or less intelligent, but three shots per second (with my computer characteristics) with a turning camera and moving adversaries, it's too small to get hired, so the camera just goes on the different side of the enemy. Here's the code:

import mss
import mss.tools
import cv2 as cv
import time
import pyautogui as pg
import numpy as np
x1, x2, x3, x4 = 0, 0, 0, 0
Conf_threshold = 0.4

NMS_threshold = 0.4

COLORS = [(0, 255, 0), (0, 0, 255), (255, 0, 0),

(255, 255, 0), (255, 0, 255), (0, 255, 255)]
class_name = []

with open('CSv3.names', 'r') as f:

class_name = [cname.strip() for cname in f.readlines()]
net = cv.dnn.readNet('CSv3_6000.weights', 'CSv3.cfg')

net.setPreferableBackend(cv.dnn.DNN_BACKEND_CUDA)

net.setPreferableTarget(cv.dnn.DNN_TARGET_CUDA_FP16)
model = cv.dnn_DetectionModel(net)

model.setInputParams(size=(416, 416), scale=1/255, swapRB=True)
with mss.mss() as sct:

monitor = {"top": 0, "left": 0, "width": 1920, "height": 1080}

output = "sct-{top}x{left}_{width}x{height}.png".format(**monitor)
starting_time = time.time()
frame_counter = 0

while True:

    sct_img = sct.grab(monitor)

    mss.tools.to_png(sct_img.rgb, sct_img.size, output=output)
    print(output)

    cap = cv.imread(output)

    x1, x2, x3, x4 = 0, 0, 0, 0

    frame = cap
    frame_counter += 1
    classes, scores, boxes = model.detect(frame, Conf_threshold, NMS_threshold)
    for (classid, score, box) in zip(classes, scores, boxes):
        color = COLORS[int(classid) % len(COLORS)]
        label = "%s : %f" % (class_name[classid[0]], score)
        cv.rectangle(frame, box, color, 1)

        x1, x2, x3, x4 = box[0], box[2], box[1], box[3]

        print (box[0], box[2], box[1], box[3])


        cv.putText(frame, label, (box[0], box[1]-10),
        cv.FONT_HERSHEY_COMPLEX, 0.3, color, 1)


    if x1 > 0:
        pg.moveTo(x1+x2/2, x3+x4/3)
    else:
        pass


    endingTime = time.time() - starting_time
    fps = frame_counter/endingTime

    cv.putText(frame, f'FPS: {fps}', (20, 50),
               cv.FONT_HERSHEY_COMPLEX, 0.7, (0, 255, 0), 2)
    cv.imshow('frame', frame)

    key = cv.waitKey(1)
    if key == ord('q'):
        break

cap.release()

cv.destroyAllWindows()

Hithab, where we can download: https://github.com/Imandyr/Auto-aim-bot-YOLOv4-python-CS-1.6

What can we do to speed it up?

UPD:
By examing, he discovered that even without the discovery of the manpower, only when he was received and displayed, the speed of work was just a little more than the discovery of the cults.

[authera]

I'd advise you to do that.

import contextlib
import time
def some_lib_func():

time.sleep(3)
@contextlib.contextmanager

def profiler_register_time_context(func_name):

start = time.perf_counter()

yield

delta = time.perf_counter() - start

print(f"{func_name} worked {delta}s")
with profiler_register_time_context("some_lib_func"):

some_lib_func()