png圖片缺乏某些庫，導致損壞，或者多餘了一些資料會導致以下報錯：

libpng warning: iCCP: known incorrect sRGB profile
libpng warning iccp extra compressed data

一些可能的解決方案：

已有方案

來自：https://blog.csdn.net/qq_37924224/article/details/119181028

libpng 1.6及以上版本增強了ICC profiles檢查，所以發出警告。從 PNG 影象中刪除 iCCP profiles。

可以先讀取，再重新儲存：

import cv2
from skimage import io
image = io.imread(path)
image = cv2.cvtColor(image, cv2.COLOR_RGBA2BGRA)
cv2.imencode('.png',image)[1].tofile(path)

 

本文提供方案

觀察到報錯的圖片根本無法開啟，選擇刪除對應圖片。

檔案結構：

- all
  - train
    - folder1
      - img1.png
      - ....
      - imgn.png
    - foler2
      - ...
  - test
    - folder1
      - img1.png
      - ....
      - imgn.png
    - foler2
      - ...

實現功能：

重新命名資料夾，重新命名檔名稱（ps: 原來是有中文的），使用convert命令轉換png為jpg

import os
import cv2 
import random 

dir_path = "all/train"


def rename_folder():
    for idx, subdir in enumerate(os.listdir(dir_path)):
        new_folder = "%s_" % (str(random.randint(0,10000))) + str(idx)
        whole_subdir = os.path.join(dir_path, subdir)
        whole_newdir = os.path.join(dir_path, new_folder)
        # print(f"rename {whole_subdir} to {whole_newdir}")
        os.rename(whole_subdir, whole_newdir)


def rename():
    # rename files
    for subdir in os.listdir(dir_path):
        whole_path = os.path.join(dir_path, subdir)
        for idx, img_name in enumerate(os.listdir(whole_path)):
            postfix = "png" if img_name.endswith("png") else "jpg"
            new_img_name = str(idx) + "." + postfix

            org_path = os.path.join(whole_path, img_name)
            new_path = os.path.join(whole_path, new_img_name)

            #     print(f"rename {org_path} to {new_path}")
            os.rename(org_path, new_path)


def convert_png2jpg():
    # convert
    for subdir in os.listdir(dir_path):
        whole_path = os.path.join(dir_path, subdir)
        for img_name in os.listdir(whole_path):
            img_whole_path = os.path.join(whole_path, img_name)
            if img_name.endswith("png"):
                # print(img_whole_path)
                img = cv2.imread(img_whole_path)
                cv2.imwrite(img_whole_path, img)
                os.system(
                    "convert %s %s"
                    % (img_whole_path, img_whole_path.replace("png", "jpg"))
                )


def main():
    # run with nohup
    # 1. rm nohup.out
    # 2. nohup python convert_png.py
    rename_folder()
    rename()
    convert_png2jpg()


if __name__ == "__main__":
    main()

 

這個時候有可能會出現報錯：

這些無法convert的檔案就是需要刪除的物件，通過正則化方法匹配對應檔案, 並刪除。

import os


def main():
    rm_list = greps()
    rm(rm_list)


def greps():
    import re

    rm_list = []
    file = "nohup.out"
    f = open(file, "r")
    for line in f.readlines():
        res = re.findall("`.+'", line)
        if res:
            rm_list.append(*res)
    return rm_list


def rm(rm_list):
    rm_list = [item.lstrip("`").rstrip("'") for item in rm_list]
    for item in rm_list:
        print("rm: ", item)
        os.system("rm %s" % item)


if __name__ == "__main__":
    main()