Knock your images before you get stressed.
One day, I had run my deep learning model training program and got off work on Friday. When I checked the training process on Monday after arriving at work, the program was terminated because of some images that cannot be opened by image processing libraries like PIL.
I developed this project to check images that cannot be opened by image processing libraries before running computer vision applications.
image-knocker is very fast and simple!
pip install image-knocker
foo@bar:~$ image-knocker --path dataset_path/ --exts png jpg jpeg Knocking... Root Path: D:\datasets\detection\VOCdevkit
100%|██████████████████| 34179/34179 [00:03<00:00, 10911.31it/s]
Corrupted image files are detected!
D:/datasets/detection/VOCdevkit/outlier.png
Corrupted image file list is saved to D:\datasets\detection\corrupted_imgs_list.txt
Knocking... Root Path: D:\datasets\detection\VOCdevkit
100%|██████████████████| 34178/34178 [00:04<00:00, 7792.30it/s]
There are no corrupted image files!
All images were loaded successfully using PIL.Image.open!