search

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
560 stars 72 forks source link

About where to find data process code. #154

Closed Yangzhouzhou closed 2 weeks ago

Yangzhouzhou commented 2 weeks ago

Hello, I have recently been studying your method. Since I haven't touched Pytorch_lighting very much before, can you tell me where is the data preprocessing department in your program?

drprojects commented 2 weeks ago

Our BaseDataset class inherits from torch-geometric and rules how raw and preprocessed data are organized.

The behavior of the datasets (and the answer to your question) is explained in the datasets documention. Please have a look.

Also, when instantiating a dataset, the logs explicitly tell you where the preprocessed data is stored.

PS: One important novelty of our BaseDataset class compared to torch-geometric is that we compute a hash to characterize the pre_transform (ie all the preprocessing steps) of a dataset. We use this hash to uniquely identify the preprocessing recipe you have used when first instantiating a dataset, which prevents recomputing / overwriting yoir preprocessed files when instantiating your dataset with a slightly different preprocessing reciper later on.