Question about Table 1 and Table 2 in paper.

[lif314]

It seems there might be a discrepancy or an error in your paper regarding the results reported in Table 1 for the ScanNet dataset and Table 2 for the TUM-RGBD dataset. The paper mentions that the results for Nice-SLAM[1] and Vox-Fusion[2] are from Nicer-SLAM [3], but I couldn't find relevant experimental results in the Nicer-SLAM paper. Additionally, the Nicer-SLAM paper only includes results for the Replica and 7-Scenes datasets.

[1] Zihan Zhu, Songyou Peng, Viktor Larsson, Weiwei Xu, Hujun Bao, Zhaopeng Cui, Martin R Oswald, and Marc Pollefeys. Nice-slam: Neural implicit scalable encoding for slam. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 12786–12796, 2022. 1, 2, 6, 7, 8, 9, 10,11. [2] Xingrui Yang, Hai Li, Hongjia Zhai, Yuhang Ming, Yuqian Liu, and Guofeng Zhang. Vox-fusion: Dense tracking and mapping with voxel-based neural implicit representation. In IEEE International Symposium on Mixed and Augmented Reality (ISMAR), pages 499–507. IEEE, 2022. 2, 6, 7, 8, 10. [3] Zihan Zhu, Songyou Peng, Viktor Larsson, Zhaopeng Cui, Martin R Oswald, Andreas Geiger, and Marc Pollefeys. Nicerslam: Neural implicit scene encoding for rgb slam. arXiv preprint arXiv:2302.03594, 2023. 2, 6.

[VladimirYugay]

Hey there, you are completely right. Thanks a lot for pointing this out.

It is a copy-paste mistake on our writing side. Rendering performance on real-world datasets (TUM_RGB, Scannet) was never reported by previous methods. We pasted this text in all rendering tables hence the error. We will fix it, thanks again!

[lif314]

Hey there, you are completely right. Thanks a lot for pointing this out.

It is a copy-paste mistake on our writing side. Rendering performance on real-world datasets (TUM_RGB, Scannet) was never reported by previous methods. We pasted this text in all rendering tables hence the error. We will fix it, thanks again!

Thank you for your work! Could you please provide the tracking results of these methods on the ScanNet and TUM-RGBD datasets? Thanks!

[VladimirYugay]

Hey there, for this work we used DROID-SLAM for our tracking backend. So if you want to track performance on those datasets you can directly check the DROID-SLAM performance in tracking.

One more thing, we are about to get permission to release the code (we hope 1 or 2 weeks at most) and plan to do it ASAP. In the released code there will be a new tracking method with new (better) results which will be also reflected on arxiv.

[lif314]

Hey there, for this work we used DROID-SLAM for our tracking backend. So if you want to track performance on those datasets you can directly check the DROID-SLAM performance in tracking.

One more thing, we are about to get permission to release the code (we hope 1 or 2 weeks at most) and plan to do it ASAP. In the released code there will be a new tracking method with new (better) results which will be also reflected on arxiv.

Thanks, looking forward to it! I will close this issue.

[zhangshuoneu]

Hey there, for this work we used DROID-SLAM for our tracking backend. So if you want to track performance on those datasets you can directly check the DROID-SLAM performance in tracking.

One more thing, we are about to get permission to release the code (we hope 1 or 2 weeks at most) and plan to do it ASAP. In the released code there will be a new tracking method with new (better) results which will be also reflected on arxiv.

Hi, Wonderful work! But I haven't noticed that droid-slam is mentioned in paper. If DROID-SLAM is used to tracking, the init camera tracking pose should be provided by droid-slam. I just find the constant speed assumption in paper. Besides, I also haven't found the droid.pth in git.

[VladimirYugay]

Hey @zhangshuoneu ,

Thanks for your interest in our work! We made another iteration of the paper to get rid of DROID-SLAM. You can find the updated version of our paper on arXiv.