Weird Mesh from provided Colab example about atlas HOT 8 OPEN

The colors represent the semantic labels (using the NYU40 colormap). So green is floor, pink is table, yellow is chair, etc.

The code also supports a model the regresses rgb color but we do not provide a pretrained model with this head.

from atlas.

That looks like the expected result. The problem is there is no shading in your visualization which makes it hard to see the geometry (also make sure to view using backface culling since Atlas often produces closed meshes). Screenshoot below with flat shading and the result from COLMAP for comparison.

Atlas:

COLMAP:

from atlas.

Thank you so much for your answer. Effectively, I had to able the shading option. But, I wonder now, why are my RGB colors so strange? In the video, the floor is kind of orange/red (carpet), but it appears green in my visualization??

from atlas.

Hey, can you please share the .ply file that you got after running inference.py. Despite doing the same execution, I'm not able to open that file, as there are no vertices generated in the .ply file.

from atlas.

Same here. And I am also wondering why is it done with --voxel_dim 208 208 80 in the readme for inference. Whereas, in the ~/atlas/config.py, the voxel configs are as below:

_C.VOXEL_DIM_TRAIN = [160,160,64]
_C.VOXEL_DIM_VAL = [256,256,96]
_C.VOXEL_DIM_TEST = [416,416,128]

Which one is correct for inference? Did you solve it? @antoinekeller Thank you very much.

from atlas.

Hi @zmurez-ml , sorry I am a bit confused. May I know is the NYU40 dataset actually the NYUv2 in ScanNet dataset? According to the NYUv2 paper, it seems like the table is not pink but orange?

from atlas.

hi, can i ask which software did you use to show .ply file? thanks

from atlas.

hi, can i ask which software did you use to show .ply file? thanks

Meshlab

from atlas.