Release artifacts (RobotDiffuse models, ROP dataset) on Hugging Face

[NielsRogge]

Hi @ACRoboT-buaa 🤗

Niels here from the open-source team at Hugging Face. I discovered your work on Arxiv and was wondering whether you would like to submit it to hf.co/papers to improve its discoverability. If you are one of the authors, you can submit it at https://huggingface.co/papers/submit.

The paper page lets people find artifacts about your work (your models and dataset for instance), you can also claim the paper as yours which will show up on your public profile at HF, and add Github and project page URLs.

It'd be great to make the checkpoints and the ROP dataset available on the 🤗 hub to improve their discoverability and visibility. We can add tags so that people find them when filtering https://huggingface.co/models and https://huggingface.co/datasets.

Uploading models

I see the code is available on GitHub. Would you like to host the pre-trained weights for RobotDiffuse on Hugging Face? See here for a guide: https://huggingface.co/docs/hub/models-uploading.

One can leverage the hf_hub_download one-liner to download a checkpoint from the hub directly into your planning scripts.

Uploading dataset

Would be awesome to make the ROP dataset available on 🤗 as well. I see you're currently using Google Drive for it. Hosting on Hugging Face will allow people to do:

from datasets import load_dataset

dataset = load_dataset("your-hf-org-or-username/ROP")

See here for a guide: https://huggingface.co/docs/datasets/loading.

Besides that, there's the dataset viewer which allows people to quickly explore the trajectories and task definitions in the browser.

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

Cheers,

Niels
ML Engineer @ HF 🤗