Let's break down exactly why you're hitting that warning and how to fix it properly—plus a smarter way to handle this without slow pixel loops.

Why the overflow happens #

• First, check your img.dtype output: it’s almost certainly uint8 (8-bit unsigned integer). This means every blue/green/red pixel value is capped between 0 and 255.
• When you do red + green + blue directly, you’re doing math on uint8 scalars first—before the result ever touches your avg variable. Even though avg is initialized as an int, the addition itself happens in uint8 space. Three maxed-out 255 values sum to 765, which is way beyond the 0-255 limit of uint8. This causes an overflow (the value wraps around modulo 256), and Python flags it with that RuntimeWarning.
• Your initial avg=1000 doesn’t help here—you immediately overwrite avg with the result of the uint8 addition, so that initial value never affects the calculation.

Why converting to int fixes it #

• When you cast each channel value to int first (int(red)+int(green)+int(blue)), you switch to arithmetic on standard Python integers, which can handle way larger values. The sum 765 fits perfectly in an int, so no overflow occurs, and the warning disappears.

A faster, more efficient approach #

• Manually looping through every pixel in Python is painfully slow for most images. OpenCV and NumPy are built for vectorized operations (processing entire arrays at once, in C-level speed). Here’s how to rewrite your code to get the same effect (setting all channels to the average of RGB values) without loops:

import cv2 as cv
import numpy as np

img = cv.imread('C:/Users/#####/Documents/roi.jpg')
print(img.shape)
print(img.dtype)

# Calculate the average of the three channels across all pixels
# Cast to int32 first to avoid overflow during summation
pixel_avg = (img.astype(np.int32).sum(axis=2)) / 3
# Convert back to uint8 to match the original image's dtype
pixel_avg = pixel_avg.astype(np.uint8)

# Assign the average value to all three channels
img[:, :, 0] = pixel_avg
img[:, :, 1] = pixel_avg
img[:, :, 2] = pixel_avg

while True:
    cv.imshow("Image", img)
    if cv.waitKey(0):
        break
cv.destroyAllWindows()

If you just want to convert the image to grayscale (which is what your code is effectively doing), you can also use OpenCV’s built-in function for even better results (it uses a weighted average that matches human perception):

gray_img = cv.cvtColor(img, cv.COLOR_BGR2GRAY)
# If you need to put it back into the 3-channel image:
img[:, :, :] = gray_img[:, :, np.newaxis]

内容的提问来源于stack exchange，提问作者user11747064