Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly

Sign up
Appearance settings

Mge/zoom #714

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

Open
ookey wants to merge 3 commits into graphdeco-inria:main
base: main
Choose a base branch
Loading
from Orange-OpenSource:mge/zoom
Open

Mge/zoom #714

ookey wants to merge 3 commits into graphdeco-inria:main from Orange-OpenSource:mge/zoom

Conversation

@ookey
Copy link

@ookey ookey commented Mar 18, 2024

Some data have image centers different from (w/2,h/2). This can be after a correction of distortion, or because some images are cropped.
To manage this case, we pass projmatrix to the rendering classes, which include cx,cy information (on top of fx, fy).

Note that this evolution depends on a change the signature of the diff-gaussian-rasterization module: CudaRasterizer::Rasterizer::forward projmatrix parameter is now the plain projection_matrix (whereas it used to be the full_projection_matrix), and tan_fovx and tan_fovy are not needed anymore.

How to test:
0. build the submodule $ pip install diff-gaussian-rasterization

  1. download dataset example with cx,cy != 0.5,0.5 https://drive.google.com/file/d/1Yg-yZy0OiZCd7P1VZrY0WtB6HJrHEIOE/view?usp=sharing
  2. train gaussian splatting on the dataset
  3. check the result with the standard sibr_player

ookey added 3 commits March 18, 2024 13:26
In some cases, calibration gives central point cx,cy != (0.5,0.5), or it
can be decided to crop the input images.
In those cases, it is necessary to split fovx to fovXleft,fovXright and fovy to fovYtop,fovYbottom
Note that the export of cameras to cameras.json merges those values back
to the basic fovx,fovy. This aims at avoiding the modification of
diff_gaussian_rasterization branch used for SIBR_gaussianViewer_app.
Signed-off-by: Matthieu Gendrin <matthieu.gendrin@orange.com>
...nsform to cuda code
full_projection_transform is the multiplication of view_matrix and
projection_matrix. Thus, passing (view_matrix, projection_matrix) is
equivalent to (view_matrix, full_projection_transform).
Passing projection_matrix as arguments to rasterize_gaussians enables
to get intrinsics informations in the rendering code. Since we'll need
central point (cx, cy), this is the aim of this refacto commit.
Signed-off-by: Matthieu Gendrin <matthieu.gendrin@orange.com>
fovx and fovy are not relevant anymore. Intrinsics are read from
projection_matrix. Thus, we don't pass them to the rendering code.
Signed-off-by: Matthieu Gendrin <matthieu.gendrin@orange.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

1 participant

AltStyle によって変換されたページ (->オリジナル) /