Finetune on ImageNet3D dataset

Hi,

Thanks for the great work! Recently I noticed there is a ImageNet3D dataset (https://github.com/Rawmantic/Dataset3D), which annoted with pose, normal and depth. I'm able to improve foundationstereo greatly with such dataset, but for depthanythingv3, I'm wondering how to finetune it with improved foundationstereo label, or use that dataset directly. Is there any training script and data preparation I can refer to?

Thanks!

No repository files, tests, or entry points are named. Start by locating the existing training and data-preparation workflows, then determine how ImageNet3D data or improved FoundationStereo labels would be represented and consumed. Done would require a working finetuning path or documented scripts for preparing and training on the requested data.

Written by the indexing model from the issue text.