search

mlvlab / SPoTr

Official pytorch implementation of "Self-positioning Point-based Transformer for Point Cloud Understanding" (CVPR 2023).
92 stars 3 forks source link

About Interpolation Feature #9

Closed docxmxm closed 7 months ago

docxmxm commented 1 year ago

Hi,

Thank you for this superior work.

I have a question about feature fi.

In code, the Interpolation feature fi was the output of cat[colors, height]. Why is this? Can I use geometry such as [coord, height] when it applies on the dataset without color? Or fi could be the SetAbstraction layer used in PointNet++?

Thank you so much for your help!

docxmxm commented 1 year ago

Or may I ask you will you pubulish other dataset parameters soon?

Rurouni-z commented 8 months ago

may i ask you what is “height” meaning？a point has height? weird to me.

i found s3dis has 4 features(r,g,b, ?), the last feature is height?

PJin0 commented 7 months ago

Thank you for the interest to our work and sorry for the late reply.

@docxmxm You can use geometry such as [coord, height] for the data without color.

@Rurouni-z The last feature of s3dis is height. We follow PointNeXt(Neurips 2023) for the data processing.