Release Mirage-Train and processed GenImage datasets on Hugging Face

[NielsRogge]

Hi @yunncheng 🤗

Niels here from the open-source team at Hugging Face. I discovered your work through Hugging Face's daily papers as yours got featured: https://huggingface.co/papers/2511.08423.

The paper page lets people discuss your work and find related artifacts. It's great to see that you've already uploaded the Mirage-Test dataset and your OmniAID model checkpoints to the Hub! You can also "claim" the paper on HF to have it show up on your profile, and add your Github link to the paper page.

It'd be awesome to make the rest of the datasets (Mirage-Train and the classified/reconstructed GenImage subsets) available on the 🤗 Hub as well, rather than just Google Drive. This would allow people to easily explore and load them via the datasets library:

from datasets import load_dataset

dataset = load_dataset("Yunncheng/Mirage-Train")

Besides easier access, there's the dataset viewer which allows people to preview the images directly in the browser.

I also noticed the models are currently .pth files in a single repository. If you're interested, you could leverage the PyTorchModelHubMixin class to add from_pretrained and push_to_hub methods to your model code, making it even easier for others to use your detector.

Let me know if you're interested or need any help regarding this!

Cheers,

Niels
ML Engineer @ HF 🤗

[O-J1]

Agree with what Niels has said, I'd like to have the option to extend it to support digital art and fine art, not just anime