You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I have tested the new code in the zip clio_ws on the apartment dataset and 00847-bCPU9suPUw9 from Habitat dataset and encountered an issue that I find puzzling.
It seems that there is an offset between the mesh of nodes extracted from the SegmentLayer and the mesh of nodes from the ObjectLayer.
Additionally, I've observed that the nodes in the SegmentLayer are aligned with the Background Mesh.
Obtaining an accurate objects node that aligns with the ground truth is crucial for my work. (Without it, I can only evaluate using nodes from the SegmentLayer, which doesn't effectively demonstrate the impact of IB.) I noticed that the code specifies a version requirement for the Spark-dsg, but setting the offset parameter to either True or False doesn't align the nodes in the ObjectLayer. And obviously setting it to False gets a wrong result as I'm using the version which is detected to be ['1', '0', '0'].(The detailed videos on these datasets are attached below.)
Have you noticed this issue, and do you have any suggestions for resolving it?
Hi Nathan,
I have tested the new code in the zip
clio_ws
on theapartment
dataset and00847-bCPU9suPUw9
from Habitat dataset and encountered an issue that I find puzzling.SegmentLayer
and the mesh of nodes from theObjectLayer
.SegmentLayer
are aligned with theBackground Mesh
.Obtaining an accurate objects node that aligns with the ground truth is crucial for my work. (Without it, I can only evaluate using nodes from the
SegmentLayer
, which doesn't effectively demonstrate the impact of IB.) I noticed that the code specifies a version requirement for theSpark-dsg
, but setting theoffset
parameter to eitherTrue
orFalse
doesn't align the nodes in theObjectLayer
. And obviously setting it toFalse
gets a wrong result as I'm using the version which is detected to be['1', '0', '0']
.(The detailed videos on these datasets are attached below.)Have you noticed this issue, and do you have any suggestions for resolving it?
Thank you for your help!
clio_zip_apartment_offset.mp4
clio_zip_00847_offset.-.Compressed.with.FlexClip.mp4
The text was updated successfully, but these errors were encountered: