Datasets:
What happened to Danbooru database & will this be updated?
This is now quite out of date compared to how up to date the Danbooru database was, but the Danbooru database is gone? What happened? Will it come back or will this continue to see updates or is there something else going on?
@Irkson0101 we are facing some urgent situation, so we have to hide these datasets.
but dont worry we are preparing for new datasets to open them, they are coming soon XD
@Irkson0101
@John6666
Dear anime enthusiasts and members of the DeepGHS community,
We are pleased to announce the optimization and re-release of more dataset options for research purposes. This includes a generally 'safe' for-all-audience dataset, appropriate for most users. We have resolved several opt-out dataset issues that were not being correctly synced in this dataset. Additionally, we have combined the previous Danbooru dataset with original entries into the main datasets. Users now have various options to access the dataset, without the need for accessing external APIs or addresses, relying on HuggingFace's fast dataset retrieval system and parquet support:
- danbooru2024-sfw, the latest SFW compliant dataset with original archives, including metadata: https://huggingface.co./datasets/deepghs/danbooru2024-sfw
- danbooru2024, the latest compliant dataset with original archives, including metadata: https://huggingface.co./datasets/deepghs/danbooru2024
- danbooru2024-webp-4Mpixel, the latest compressed compliant dataset with images compressed to no more than 4M pixels and stored in webp format, including metadata: https://huggingface.co./datasets/deepghs/danbooru2024-webp-4Mpixel
We emphasize here that we will continue our dedication to better datasets, including training custom-aesthetic reward scoring models, which resolves several issues in other models. Large datasets are mandatory, and we are getting closer and closer to a true open-source AI ecosystem in all aspects. We will try to maintain an open stance for all data types, however with proper measures as mentioned.
We are optimistic about the future and plan to reopen the API as a bot or some platform on HuggingFace, which can be used for DPO/preference optimization in the future. If you have any ideas to prevent a free lunch, please share them.
Stay tuned for our future updates and progress!
Hello. Sorry. I didn't notice it 9 days ago, but I've noticed it now.😅 I don't have any particular plans, but I just wanted to say hello.