Datasets:

zh-plus

/

tiny-imagenet

like 63

Hi,

I have encountered an issue while using the Tiny ImageNet dataset. It appears that a small portion of the images are grayscale instead of RGB, even though the dataset is expected to contain only color images (RGB).

Here are the details:

• Dataset: Tiny ImageNet
• Issue: Some images are in grayscale (1 channel), while the majority are in RGB (3 channels).
• Expected Behavior: All images should be in RGB format (3 channels).
• Actual Behavior: A few images have only 1 channel (grayscale).
• Impact: This causes issues during model training and visualization, as these grayscale images do not match the expected RGB input format.

Here is one example: the 1854th image

<PIL.JpegImagePlugin.JpegImageFile image mode=L size=64x64 at 0x7F38C81F3290>

Would it be possible to ensure that all images in the dataset are converted to RGB format, or provide a clear list of grayscale images so that users can handle them appropriately?

Thanks for your attention to this issue.

Best regards.