Question about training dataset

[piddnad]

Thanks for your great work!

I read the paper and found that the following sentence was mentioned in the experiment section.
"We train our model on 1.2M color images of the ImageNet 1K training set after excluding 10% original images with low colorful- ness scores."

Which pictures were removed from the original ImageNet training set?
I would appreciate it if you could share a list of deleted images or code to detect low-quality images.

[KIMGEONUNG]

Thank you for your interest on our research.

I added a file you want to get into this repository (here)

More specifically, we excluded the images with a less colorfulness score than 15, which is the lowest colorful category defined in a research "Measuring Colourfulness in Natural Images".

If you need the measurement code, refer here

If you have any remaining question, please let me know.

[piddnad]

Your reply is very clear and answers all my questions.

Thank you for your kindly reply!

[AlexanderKozhevin]

@piddnad sorry, were you able to run inference with pretrained model?