Would it be possible to provide a downloadable link for the rendered images? about zero123

[ruoshiliu]

Hi @MultiPath , thank you for your interest in our work. We understand the difficulty of rendering the dataset. We are working with huggingface to host our training dataset there for people to download. Stay tuned for the updates the in the next few days!

[ruoshiliu]

We used 10 machines (each with two 20-core CPU and 8x2080Ti GPU) to render for around 5 days.

[claforte]

Was this using EEVEE or Cycles?

Nevermind, I see in the paper you used Cycles. Thanks for the great paper and dataset!

[ruoshiliu]

Dataset can be downloaded here:
https://github.com/cvlab-columbia/zero123#objaverse-rendering-preliminary

P.S. This is a temporary solution so the downloading speed can be very slow and can take several days to finish.

[MultiPath]

Thanks! ETA 12h is acceptable!! Really appreciate it.

[ruoshiliu]

@mitming , we'll try to keep the link active until we find an alternative way of hosting.

[ruoshiliu]

Looks like the link has been working as expected -- closing this issue.

[MultiPath]

Hi, that would be awesome!! Looking forward to running on the same settings soon!

[claforte]

Thanks a lot in advance @ruoshiliu for sharing that dataset! Like @MultiPath , I'm also interested in this question:

Do you have an estimation of how long the whole process would take to render the objaverse dataset?

[claforte]

Thanks @ruoshiliu ! Was this using EEVEE or Cycles?

[MultiPath]

Just wondering any progress on this. Really appreciate it in advance. @ruoshiliu
I have tested rendering by myself, but I found the speed is much slower than what you showed above.
Will the number of workers affect the rendering speed (I tried 8)? If the data is available soon then I would be super grateful.
Thanks again

[MultiPath]

Just to clarify, the problem on my side using CYCLE seems it always showing to be "loading rendering kernel" for each camera view and it is extremely slow.

[ruoshiliu]

Still working on the dataset. It could take a few more days. I'll update under this thread when it's ready for download.

The total number of workers was 16 (workers_per_gpu=2) for me and anything beyond that didn't make any difference. Too many workers could put unnecessary load to CPU.

[ruoshiliu]

Hi all, I'm sorry to let you know that we might need to close downloading of the dataset soon due to higher-than-expected traffic, which is causing some difficulty to our cluster. We are seeking a different solution and will provide updates shortly. It will remain open for the next 24 hours but no guarantee after that. Sorry for the inconvenience.

[liming-ai]

Hi all, I'm sorry to let you know that we might need to close downloading of the dataset soon due to higher-than-expected traffic, which is causing some difficulty to our cluster. We are seeking a different solution and will provide updates shortly. It will remain open for the next 24 hours but no guarantee after that. Sorry for the inconvenience.

Hi @ruoshiliu,

May I ask if it can be postponed for a period of time, I still need about 40 hours to download all the data, and I have downloaded 500G so far, it would be a pity for me to just give up. Please remain open for 48 hours, thank you very much!

[liming-ai]

@mitming , we'll try to keep the link active until we find an alternative way of hosting.

Thanks for your effort!

[ruoshiliu]

Hi all, the dataset hosted on AWS S3 can be downloaded from here:

wget https://tri-ml-public.s3.amazonaws.com/datasets/views_release.tar.gz

Please let me know if it works.