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
546 stars 71 forks source link

s3dis dataset corrupt #54

Closed hansoogithub closed 8 months ago

hansoogithub commented 8 months ago

when i downloaded the Stanford3dDataset_v1.2.zip this file in line 323473 has a symbol that makes it corrupted Area_5/office_19/Annotations/ceiling_1.txt

when i train using python src/train.py experiment=s3dis datamodule.xy_tiling=2 datamodule.fold=6 it runs fine

python src/eval.py experiment=s3dis datamodule.xy_tiling=2 datamodule.fold=6 ckpt_path=mycheckpoint it cant run spt_error.txt

could this data corruption cause the error?

drprojects commented 8 months ago

Hi. I have not encountered any corrupt data error with the Stanford3dDataset_v1.2.zip data, but I know that some people have build patches for some files (see here). I also found some errors, you can find my personal patches applied when you read raw S3DIS data here.

Yet, I have not encountered any issue with Area_5/office_19/Annotations/ceiling_1.txt.

The error you are encountering is basically saying that your cloud is empty. You will need to manually inspect what comes out of your data reader (ie read_s3dis_room and read_s3dis_area). See if any room produces an empty point cloud and why.

drprojects commented 8 months ago

Have you looked into this issues and fixed it ? May I close it ?

hansoogithub commented 8 months ago

yes place close thank you

samira-jpg commented 1 month ago

May I ask how this error was fixed? I'm encountering the same error!