Request: Older PaGeR models/repos.

#2
by ApacheOne - opened

Hello,
The older PaGeR model repos are not public anymore. I am revisiting these models and would be grateful to have access to them again so my comfyUI nodes that use the older models can use them. Apache0ne/Comfyui-marigold-intrinsics-PaGeR and Apache0ne/ComfyUI-marigold-intrinsics-PaGeR-TensorRT

The repos names:
-prs-eth/PaGeR-depth
-prs-eth/PaGeR-metric-depth (the older version)
-prs-eth/PaGeR-depth-indoor
-prs-eth/PaGeR-metric-depth-indoor
-prs-eth/PaGeR-normals-Structured3D

If you need to contact me elsewhere to make this possible, here is my student email: ks264624@ohio.edu
Thank you!

Photogrammetry and Remote Sensing Lab of ETH Zurich org

Hi,
Unfortunately, we are not planning to bring the old Pager models back, as the current ones are related to and properly referenced in our paper.

vulus98 changed discussion status to closed

by using https://hanshengchen.com/asymflow/ methods and the older sd2.1 based architecture, I'd like to try implementing the latent to pixel space training so that we get better output because the diffusion architecture allows to generate higher detailed outputs over the currently provided model that does not have the ability to add noise for more detail.
I realize diffusing new details from noise may diverge from the accurate depth intent of pager but it has applications for higher fidelity in other areas.
I would really appreciate if you could provide me with the older version to expand on the research, thanks for your time and pager work.

Photogrammetry and Remote Sensing Lab of ETH Zurich org

Hi,
I could share the models to you (via google drive or smth like that), but I don't plan to publish them on HF. If you provide me with your email, I could send you the link to download them and invite you to the private repo to run them, if needed. The thing is, the paper with the old diffusion backbone is not published, so technically there is no paper to cite there, just so you are aware of that

Thank you, my email is: xcoolcat502@gmail.com
A Google drive would be great and the invite to the private repo.
I can reference the models as unpublished models in my ComfyUI node work.

Sign up or log in to comment