about dataset

[huiquandeng]

Where I can get the training data constructed by researchers based on the Objaverse dataset？

[iamNCJ]

Hi, thanks for your interest in our work. Unfortunately, we can’t release the original training data, as they haven’t yet cleared company-wide compliance.

All training data preparation details are fully described in the paper, and the data structure is identical to what scene_processor/to_h5.py currently produces, with just one extra field per sample: an HDR ground-truth image rendered via Blender.

If you hit any implementation snags while reproducing the results, please don't hesitate to contact us.

[huiquandeng]

It's really generous of you to offer assistance with implementation questions. Thanks for taking the time to provide these clarifications!  zphyrian ***@***.***  

…

------------------ 原始邮件 ------------------ 发件人: "Chong ***@***.***>; 发送时间: 2025年7月17日(星期四) 中午11:39 收件人: ***@***.***>; 抄送: ***@***.***>; ***@***.***>; 主题: Re: [microsoft/renderformer] about dataset (Issue #9) iamNCJ left a comment (microsoft/renderformer#9) Hi, thanks for your interest in our work. Unfortunately, we can’t release the original training data, as they haven’t yet cleared company-wide compliance. All training data preparation details are fully described in the paper, and the data structure is identical to what scene_processor/to_h5.py currently produces, with just one extra field per sample: an HDR ground-truth image rendered via Blender. If you hit any implementation snags while reproducing the results, please don't hesitate to contact us. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: ***@***.***>