Datasets:

Languages:
English
ArXiv:
License:

The new XL dataset link

#38
by nand-tmp - opened

I read the paper https://arxiv.org/pdf/2212.08051.pdf where is the new dataset link?

nand-tmp changed discussion status to closed
nand-tmp changed discussion status to open

same problem.

Would be friendly to explain the situation, as this was promised for end of June ...
Is there a new release date people can hope for?

Allen Institute for AI org

Oops, sorry, forgot I said that in the paper πŸ™ˆ. More like end of July. The data is ready, just working on some details involving the release. I will try to get sooner though if possible.

In general, my recommendation is to use Objaverse 1.0 for prototyping. If your idea works with Objaverse 1.0, it can then be scaled up further to Objaverse-XL, and should hopefully follow somewhat predictable scaling curves (as we show for experiments in our paper). I probably wouldn't start with Objaverse-XL without doing smaller scale experiments first.

Allen Institute for AI org

I'll update with a link here when the Hugging Face repo is public :)

Thanks very much for the quick answer! This helps. Another question: will there ever be a possibility of tuning your model towards a personal touch based on a small add-on set of meshes, something like creating a Hypernetwork in SD?

Allen Institute for AI org

Yeah, definitely possible!

There's already some work along these lines, which doesn't even need meshes, just images:
https://dreambooth3d.github.io/

But it's still an active area of research and I'm hopeful a 3D equivalent of HyperDreamBooth will be out soon :)

So the plan is to give the data as python API again or some other method ?

Awesome! Looking forward to download the full model soon and try out new things : )! can't wait for it to be downloadable in full in a suitable format.

Really thank you for this great dataset. Just curious to know if there are any updates or estimate release dates for the XL version of the dataset. Thank you so much again!

Allen Institute for AI org

Sorry for the delay, we've discussed internally and with others and planned to release everything next week!

Sounds as convincing as exiting! What a commitment, thank you very much.

Sorry for the delay, we've discussed internally and with others and planned to release everything next week!

@mattdeitke thank you very much : )! I've even set a reminder for this week to check haha. Any updates on that :)?

rather admirable ability hiding 10 million things! Some kind of antimatter, probably.

Allen Institute for AI org

Starting to upload today, bear with me while I upload things πŸ˜„

Here's where things will go! I'll also add in an API to facilitate downloading things and provide the updated Blender rendering code that supports all these new file formats:
https://huggingface.co/datasets/allenai/objaverse-xl

Many many thanks, looking forward to this very much!

Starting to upload today, bear with me while I upload things πŸ˜„

Here's where things will go! I'll also add in an API to facilitate downloading things and provide the updated Blender rendering code that supports all these new file formats:
https://huggingface.co/datasets/allenai/objaverse-xl

I can only imagine how long this upload will take O.O! @mattdeitke thank you : )!

excited about this upload!

I greatly appreciate your dedication in curating this dataset. If feasible, could you please provide an estimated timeline for the complete release of the entire dataset?

@mattdeitke Is the upload finished :)? I would love to use the new dataset's great potential : )! I saw the temporary notice of "currently uploading dataset" for a while, but it seems to be gone : )

Let me join in on this, as I am looking daily for news about upload related progress. I have announced several presentations and talks and workshops to explore the potential of your achievement, but now I feel a bit left alone. So any news would be good news.

Allen Institute for AI org

Hey, sorry for the delay! I have things ready for GitHub, just need the final sign off from the team. I will try my best to get that done tomorrow, as I'm trying to release asap.

The initial push of the Thingiverse and the Smithsonian data is done. Here is the API for downloading (https://huggingface.co/datasets/allenai/objaverse-xl/tree/main/objaverse_xl). I've also added Objaverse 1.0 (the original Sketchfab data) to this same API and updated the code quite a bit to make it cleaner and easier to specify the download directly (and. download directly to the cloud if desired).

Polycam likely within a week.

Hope that helps! Sorry for the delay, I've been trying to push it out!

Allen Institute for AI org

Happy to help answer questions about downloading as well if there are any issues or questions!

Allen Institute for AI org
β€’
edited Aug 25, 2023

Also, haven't linked to this anywhere, but the Zero123-XL model is public here, and so is the code for the Dreamfusion fine-tuning stage to get the image-to-3D and text-to-3D results with threestudio!

Oh this is all great and useful, thanks very much for your immediate answers. As I am more the kind of designer type using mainly OS X and Windows, I was glad to understand that you would be preparing some kind of blender add-on to use the Zero123-XL model. Is this correct, and if so, when do you think that this will be available?

Happy to help answer questions about downloading as well if there are any issues or questions!

Hello, I would like to know how much memory space is required to download the 3D data and rendering data for objaverse and objaverse-XL?

Allen Institute for AI org

For Objaverse-XL, around 17 TB for rendering (12 views each of 512x512)

I’m not sure on object storage.

Sign up or log in to comment