feature of the different pose sample1 pcd

[Jar-17]

Dear Concerto team,

Thanks for the amazing work! I am writing to inquire about the feature of the different pose sample1 pcd I observed

I ran an experiment on sample1 pcd and then turned it 90degrees along the y-axis, as shown

When I run the demo code without color and normal for these two PCDs, the PCA result is as below

The color for the same part is different in these two PCDS, which, to my knowledge, means they have different features. Is that the property of Concerto, or do I need to adjust something when changing the pose of pcd?

Thanks for your great work again!

[Yujia-Zhang0913]

Hi,
Concerto is focused on scene-level data. We assume that all data is z-axis up with small x y rotations as augmentations. Therefore, the features of your y-rotated data are different from the original one.
However, you can try rotating in z axis and input the original features and the features from the rotated one to the PCA process at the same time(Since PCA has random parts every time). The visualization should be the same. Here is my implementation for z rotation and shift:

And we currently try to erase the gravity alignment for the objects in the scene, which is a part of our next work, Utonia.