Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

What is the use of Render Flow function ? #43

Open
sauradip opened this issue Jun 6, 2024 · 1 comment
Open

What is the use of Render Flow function ? #43

sauradip opened this issue Jun 6, 2024 · 1 comment

Comments

@sauradip
Copy link

sauradip commented Jun 6, 2024

Hi,

I observed Render Flow method in

loss = loss + lambda_optical * optical_flow_loss
, and how is this 2D optical flow being projected back to 3D space ? I did not find this in the paper !

@yihua7
Copy link
Owner

yihua7 commented Jun 6, 2024

Hello,

  • The render_flow function related to what you mentioned is indeed optional and not typically used; it is primarily useful for flow supervision when RAFT flow or FlowFormer flow data is available.
  • Rendered 2D optical flow would not be projected back to 3D but can be supervised by estimated optical flow.
  • While this feature remains unhighlighted in our paper, we have made the implementation accessible in the hope that it might benefit those in the community who require it. Although another paper, GaussianFlow makes it one of the main contributions, we hold that the point is too tiny to build a contribution.
  • Interestingly, some features like the dynamic mask and embedding for dynamic Gaussians are highlighted as significant innovations in the papers GauFRe and Per-Gaussian Embedding-Based Deformation for Deformable 3D Gaussian Splatting. In contrast, these components were already utilized in our SC-GS framework but were considered too minor to be emphasized separately in our publications. A deeper look into our code would reveal that these are essentially the dynamic mask and hyper coordinate parameters.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants