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

PerspectiveCameras principal point #1276

Open
classner opened this issue Jul 31, 2022 · 1 comment
Open

PerspectiveCameras principal point #1276

classner opened this issue Jul 31, 2022 · 1 comment
Assignees
Labels
question Further information is requested

Comments

@classner
Copy link
Contributor

🐛 Bugs / Unexpected behaviors

Hi team!

Someone just reached out to me concerning unexpected behavior of PerspectiveCameras in conjunction with Pulsar. I tracked this back to here

principal_point=((0.0, 0.0),),
. The default initialization for PerspectiveCameras uses a principal point of (0., 0.), which, according to the documentation is the xy coordinate of the principal point (and hence places it in the top left corner). Is that intended and shouldn't it be half of image size by default? Could it be that this definition changed during the last year at some point? In any case, the Pulsar unified interface is currently using this value incorrectly in and . I'm happy to make a PR once clarified how it's intended to work.

@gkioxari gkioxari self-assigned this Aug 10, 2022
@gkioxari
Copy link
Contributor

@classner
This is NDC coordinates, so (0,0) is the center of the image.

@gkioxari gkioxari added the question Further information is requested label Aug 10, 2022
classner added a commit to classner/pytorch3d that referenced this issue Oct 26, 2022
This commit addresses several braking changes to PyTorch3d camera conventions that made the unified system unusable at the moment (see facebookresearch#1352 and facebookresearch#1276). This commit fixes all these problems and tests using the test case from facebookresearch#1352. The results are equivalent for `FoVOrthographicCameras`, `OrthographicCameras`, `FoVPerspectiveCameras` and `PerspectiveCameras`.

We used to have a test case for the unified interface for Pulsar that compared the rendering results with some 'ground truth' images. This test seems to have been removed, thus leading to these breaking changes remaining undiscovered for a long time. Would be great if we could revive that test case!
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
question Further information is requested
Projects
None yet
Development

No branches or pull requests

2 participants