Surface Normals Edge "Artifacts"

[KevinQu7]

Thanks for this great dataset!

I am currently working with the surface normals provided for the “Small City” dataset and have noticed some strange behavior at the boundaries where sky and non-sky pixels touch (see the first three attached images for an example). After normalizing the vectors to unit length, there is a green line that runs continuously across the boundaries.

When inspecting the pixel values in the raw .exr file (see 4th attached image), a (bilinear) interpolation seems to have happened at the edges between the sky pixels (0-vector) and the non-sky pixels, which then causes these artifacts after normalizing. Unfortunately, these artifacts can be observed in almost every surface normal image in the dataset (every time the sky is visible).

Would it be possible for you to upload the surface normals for the "Small City" which don't have these interpolation artifacts?

Thank you very much!

[yixuanli98]

Can you tell me the image you visualize in the first image?

[KevinQu7]

The image is small_city_road_outside/0009

[Binyr]

Hi, in your first normal image, the ground is blue. But when I loaded normal and visulize it by "(n*0.5 + 0.5)*255", the ground is pink. How do you visualize the normal?