drprojects / superpoint_transformer

Official PyTorch implementation of Superpoint Transformer introduced in [ICCV'23] "Efficient 3D Semantic Segmentation with Superpoint Transformer" and SuperCluster introduced in [3DV'24 Oral] "Scalable 3D Panoptic Segmentation As Superpoint Graph Clustering"
MIT License
508 stars 65 forks source link

Point cloud with intensity and without RGB. #130

Closed zeejja closed 2 weeks ago

zeejja commented 2 weeks ago

Greetings,

First of all, thank you very much for the effort put into this repository and the articles. It's really impressive. I just have a small question before conducting the experiment with my personalized data. I want to know if I can use it on my custom point cloud data that only contains XYZ + intensity information without RGB information, or on data that includes both RGB and intensity information. Thank you for your effort, I am looking forward to your response.

drprojects commented 2 weeks ago

Hi @zeejja, thanks for your interest in the project !

Yes of course you can use any combinations of {XYZ, intensity, RGB}. Have a look at our already-implemented datasets and the corresponding configs to see how to handle your point attributes. For instance, DALES has XYZ+intensity, while S3DIS, ScanNet, and KITTI-360 have XYZ+RGB. None of those actually use both intensity and RGB, but it would be possible configure your dataset for it. To this end, have a look at the partition_hf and point_hf in the datamodule configs:

As long as your dataset's read_single_raw_cloud() returns a Data object with rgb and intensity attributes, implementing what you want should be straightforward.

Happy coding !