About fine-tuned model on Scannet.

[ootts]

Thanks for your nice work! According to your paper, you have fine-tuned the baseline methods MVDepthNet, GPMVS, and DPSNet on the ScanNet dataset. I wonder if you could share the fine-tuned models with us? I'm working on a follow-up work of Atlas and I'm making some modifications to the depth map filtering and fusion pipeline so that I will need to re-run all the TSDF integration and evaluation metrics. Sharing the predicted depth maps on the ScanNet will also suit my needs, however, the files to share will be in a much larger size. Thanks a lot!

[zmurez]

Unfortunately I do not have that code to share. But we just used the respective author's code and wrote a scannet dataset for them. It would be good for someone to implement a unified trainer with each of these methods implemented for the best fair comparison (same data augmentation, training schedule, etc.).

[ootts]

Actually, what we need is the trained models or checkpoints for the baseline methods MVDepthNet, GPMVS, and DPSNet. It will be great if you can share them with us.

[zmurez]

Unfortunately, I also no longer have access to the checkpoints (I no longer work at Magic Leap).