Help for Usage of CUDA

I have a code that I process images with a dataset from “Roboflow Train”. With this code, I saw low fps values so I need to change this code with the OpenCV CUDA module but I don’t really know about the module. So can you please help me how can I change this code with CUDA module?

import base64
import cv2
import numpy as np
import requests
import time

# --- constants ---

ROBOFLOW_MODEL = 'isas-22-kirmizi-plaka/2'

# --- functions ---

params = {
    "api_key": ROBOFLOW_API_KEY,
    "format": "image",
    "stroke": "5"

headers = {
    "Content-Type": "application/x-www-form-urlencoded"


def infer(img):
    # Resize (while maintaining the aspect ratio) to improve speed and save bandwidth
    height, width, channels = img.shape
    scale = ROBOFLOW_SIZE / max(height, width)
    img = cv2.resize(img, (round(scale * width), round(scale * height)))

    # Encode image to base64 string
    retval, buffer = cv2.imencode('.jpg', img)
    img_str = base64.b64encode(buffer)

    # Get prediction from Roboflow Infer API
    response =, params=params, data=img_str, headers=headers, stream=True)
    data =


    if not response.ok:
        print('status:', response.status_code)
        print('data:', data)

    # Parse result image
    image = np.asarray(bytearray(data), dtype="uint8")
    image = cv2.imdecode(image, cv2.IMREAD_COLOR)

    return image

# --- main ---

video = cv2.VideoCapture(0)

while True:
    start = time.time()

    ret, img =

    if ret:

        image = infer(img)

        if image is not None:
            cv2.imshow('image', image)

            if cv2.waitKey(1) == ord(
                    'q'):  # `waitKey` should be after `imshow()` - to update image in window and to get key from window

            end = time.time()
            print(1 / (end - start),
                  "fps")  # print() automatically add space between elements - you don't need space in "fps"

# - end -

Hi @caganoztopal55 - here is the Roboflow Documentation for deployment to NVIDIA Jetson devices: NVIDIA Jetson (On Device) - Roboflow

Thanks, but this code will not go on Jetson. It will run on a computer.

Additionally, there is an Enterprise option that runs at a higher FPS: Enterprise Deployment - Roboflow

Or you can also try running the Hosted Inference API code in async mode