libpng - RGB color space not permitted on grayscale PNG when using images created by Pillow

Question

I am trying to train a YOLOv8 detection model. Since I couldn't find a dataset that fits my project's purpose, I've decided to create my own dataset by overlaying images on top of a background image. To achieve this, I used Python's PIL package. However, a few errors consistently show up when training the YOLO model.

Below is the code I used to generate images.

from PIL import Image, ImageCms
from glob import glob
from tqdm import tqdm

import numpy as np
import os
import shutil
import random
import sys

def overlay_and_resize_pillow(background_image_path, overlay_image_path, output_image_path, white):
    # Open the background and overlay images
    if white:
        background = Image.new(mode="RGBA", size=(random.randint(512, 1024), random.randint(512, 1024)), color="white")
    else:
        background = Image.open(background_image_path).resize((random.randint(512, 1024), random.randint(512, 1024)))
    overlay = Image.open(overlay_image_path)
    
    profile = ImageCms.createProfile("sRGB")

    # Generate a random scale factor for resizing the overlay image
    scale_factor = random.uniform(0.5, 1.2)

    # Resize the overlay image based on the scale factor
    overlay_resized = overlay.resize((int(overlay.width * scale_factor), int(overlay.height * scale_factor)))

    # Get the dimensions of the resized overlay image
    overlay_width, overlay_height = overlay_resized.size

    # Calculate the valid range for overlay position to ensure it fits within the background
    max_x_offset = background.width - overlay_width
    max_y_offset = background.height - overlay_height

    # Calculate the position to overlay the resized image on the background
    x_offset = random.randint(0, max_x_offset)
    y_offset = random.randint(0, max_y_offset)

    # Create a copy of the background to overlay the resized image
    background_copy = background.copy()

    # Paste the resized overlay image onto the background copy
    background_copy.paste(overlay_resized, (x_offset, y_offset), overlay_resized)

    # Save the output image
    background_copy.save(output_image_path, icc_profile=ImageCms.ImageCmsProfile(profile).tobytes())
#     background_copy.save(output_image_path)
    
    name = output_image_path.split('/')[-1].split('.')[0]
    
    cl = '0'
    width = overlay_width / background.width
    height = overlay_height / background.height
    
    x_center = (x_offset + overlay_width / 2) / background.width
    y_center = (y_offset + overlay_height / 2) / background.height
    
    with open(f'./chartdetection/labels/train/{name}.txt', 'w+') as f:
        f.write(' '.join(map(str, [cl, x_center, y_center, width, height])))

When training, I keep getting these errors

• libpng warning: iCCP: profile 'ICC Profile': 'RGB ': RGB color space not permitted on grayscale PNG
• libpng warning: iCCP: known incorrect sRGB profile

How should I modify my code?

Answer 1

Convert your image to RGB before trying to save it with a color profile:

    background_copy = background_copy.convert("RGB")
    # Save the output image
    ...