As a feature extractor for point cloud

[GentlesJan]

Hi,
Is your pre-trained model suitable as a feature extractor for a point cloud shape, similar to PointNet? If it is suitable, which variable in the code should I output?
I have observed that the shape of GPT_extractor is [batchsize, 64, 1024], which seems to be the features of the points group. Are there any features based on individual points?
Thanks!

[CGuangyan-BIT]

I haven't yet come up with a good way to directly map the token features extracted by PointGPT back to point features. However, perhaps they could be used as global features, and then another network could be used to extract per-point features. Adding the two together would leverage PointGPT's pre-trained network parameters, which should improve the ability to extract features for each point.