MoGe icon indicating copy to clipboard operation
MoGe copied to clipboard

Request for ROE Solver Implementation

Open while0l1 opened this issue 1 year ago • 2 comments

Thank you for your excellent work on MoGe! I'm currently fine-tuning the model using my own data, and while the training code is progressing well, I find that aligning the predicted pointmap with the ground truth during training is an important step.

I have been trying to achieve this alignment based on Equation 4 and Algorithm 1 in your paper. However, my implementation of the ROE solver has not yielded satisfactory results. Would it be possible for you to share the implementation of the ROE solver mentioned in the paper? It would be incredibly helpful for my experiments.

Thank you for your time and contributions!

while0l1 avatar Dec 06 '24 16:12 while0l1

Hi. Thank you for your interest in our research!

Our team is currently discussing the release plan. We aim to release the training and evaluation code within the next few weeks. If possible, we plan to make the alignment code available first to facilitate your implementation. We hope you understand our need to protect the originality of our research while the work is still in its pre-publication stage.

In the meantime, for your own implementation, you may find more detailed descriptions of the solver algorithm in the supplementary materials (Algorithms 2–5). I hope this helps!

EasternJournalist avatar Dec 08 '24 07:12 EasternJournalist

hello, and thanks for your excellent work! Do you plan to release the ROE solver in the coming 1~2 weeks?

thanks again for the update!

ruili3 avatar Jan 14 '25 07:01 ruili3