ImageAI 目标检测非常慢

Question

I have been attempting to make a program that can identify objects on my screen using python. With the help of this tutorial: https://stackabuse.com/object-detection-with-imageai-in-python , I have created the following code.

import pyautogui
import cv2
import time
import os

os.add_dll_directory("C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v11.4/bin")

from imageai.Detection import ObjectDetection
import tensorflow as tf

detector = ObjectDetection()

model_path = "./models/yolo-tiny.h5"
input_path = "./input/test45.jpg"
output_path = "./output/newimage.jpg"

detector.setModelTypeAsTinyYOLOv3()
detector.setModelPath(model_path)
detector.loadModel(detection_speed="fastest")
while True:
    start_time = time.time()
    x = np.array(pyautogui.screenshot())
    x = cv2.cvtColor(x, cv2.COLOR_BGR2RGB)
    y, detection = detector.detectObjectsFromImage(input_type="array", input_image=x, output_type="array")
    for eachItem in detection:
        print(eachItem["name"] , " : ", eachItem["percentage_probability"])
    print("FPS: ", 1.0 / (time.time() - start_time))

I would like to be able to achieve 20-30 fps, however, I can only get around 1 fps and I do not know what is slowing it down.

My specs are:

• 2 x RTX 3060
• Intel i7 4770
• 24GB RAM

All help is greatly appreciated.

Answer 1

Is the program using your CPU or GPU? If it uses the CPU, you might wanna check that you have the right version of tensorflow-gpu installed (if at all). ImageAI will prefer your GPU if it has access to it, but 20-30fps is pretty much impossible with it - too inefficient.