Frédéric Devernay

IMHO this should become the default, because: - "antialiased" hold a closer-to-reality physical representation of the scene - it doesn't hurt eval results when eval images are at the same...

The fact that third-party viewers don't render 2D Gaussians correctly is their own issue (some code don't even have the +0.3 trick)

Do you mean that kind of artifacts? https://github.com/nerfstudio-project/nerfstudio/issues/1498 If what you want to do is training with transparency (transparent areas should have no matter), and not training with masks (areas...

The fix should be done upstream: w and h returned by the latest version of icvGetRectangles are off by 1 pixel, see https://github.com/opencv/opencv/issues/24831

I would recommend adding a unit test to test_datamanager.py, so that we're sure to get the thing right in the end.

And btw we should probably not use cv.undistort, which uses bilinear interpolation, and write our own function that calls cv2.remap with INTER_CUBIC instead. bilinear interpolation creates uneven blur in the...

> Any suggestions what we should do with this issue with the cropped pixel? gate the fix you did with if 4.8

First of all, if your vertical is not right, I would try `--orientation-method vertical` which fixes many issues with the default method (`--orientation-method up`), see https://github.com/nerfstudio-project/nerfstudio/issues/1765 If you implement another...

This is an option to ns-render, not ns-train. Here's what you would use to render masks (aka accumulation): ``` ns-render ... \ --output-format images --image-format png --rendered-output-names accumulation \ --colormap-options.colormap...

Ok so here are my explanations. First of all, calibrating long focal length with COLMAP may give you a good reprojection error, but the actual optical center of the camera...