Demo video

[mix345]

Thank you very much for your work and the code release! The result is pretty cool.

I have a question about the demo video in the supplementary material. In the video, an accurate depth map is rendered from the first frame of the sequence (before starting the process). I am confused because such a depth map can be generated after processing multiple frames and optimizing the implicit map via multi-view posed images. Could you tell me the reason?

Thanks!

[poptree]

Hi,

As we mentioned in the paper, we use 15-20 frames to generate the initial depth map in the initialization progress. In our code, we provide the implementation of the initialization. We can see both the depth map and pose are optimized without prior.

I am confused because such a depth map can be generated after processing multiple frames and optimizing the implicit map via multi-view posed images.

Correct, but what the relationship between this claim and our method?

[mix345]

Thanks for the reply. OK, so does the rendered depth at the beginning of the video show the one after the initialization process? Then it makes sense.

[poptree]

Thanks for the reply. OK, so does the rendered depth at the beginning of the video show the one after the initialization process? Then it makes sense.

Yes.