Usage of Depth Maps

[JesseZZZZZ]

Hi, dear authors!
Thank you for your great work, I have a small issue. In your paper, the depth information is part of the patch. But during inference, the depth maps are initialized all zero, how can this be? Is this because depth is only related to optical flow loss, and not related to trajectory loss, so you don't need depth maps during inference?

Thank you!

[lushouyi]

The author did not use the ground truth depth information during the training process. Instead, the depth values were set to random numbers.

[JesseZZZZZ]

@lushouyi Are you sure...? I believe the disps is the inverse depth... Only the depth map durence inference is initialized all zeros

[lushouyi]

Line 206 of the net.py file

[JesseZZZZZ]

Thanks，seems like the newest version is a bit different...

…

---- Replied Message ---- | From | Shouyi ***@***.***> | | Date | 12/03/2024 21:50 | | To | ***@***.***> | | Cc | ***@***.***>***@***.***> | | Subject | Re: [princeton-vl/DPVO] Usage of Depth Maps (Issue #88) | Line 206 of the net.py file — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: ***@***.***>

[JesseZZZZZ]

@lushouyi Well, I don't understand every line of code, but I think the depth map is used somewhere, because NAN appears when I change the depth maps to all zeros

[lushouyi]

You can set everything to 1 instead, I feel that setting everything to 0 might be problematic.