s3dis preprocess
Closed this issue · 6 comments
When preprocess Stanford3dDataset_v1.2_Aligned_Version using python pcr/datasets/preprocessing/s3dis/preprocess_s3dis.py --dataset_root ${RAW_S3DIS_DIR} --output_root ${PROCESSED_S3DIS_DIR}, i meet an error
Traceback (most recent call last): File "pcr/datasets/preprocessing/s3dis/preprocess_s3dis.py", line 114, in <module> main_process() File "pcr/datasets/preprocessing/s3dis/preprocess_s3dis.py", line 93, in main_process if "Aligned_Version" in opt.source_root: AttributeError: 'Namespace' object has no attribute 'source_root'
Hi, thanks for your kind reminder and I have fixed the bug. Also, you can try our new advanced preprocessing script for S3DIS in Pointcept.
Thank you for your reply.
I have noticed a new data processing script for S3DIS in pointcept, and I would like to know if ptv2 will perform better when using the new data processing？Can you briefly explain the differences between the new way of data processing and the previous one？
If I download Stanford 3dDataset_ v1.2_ Aligned_ Version.zip, do I only need to add "-- align_angle" to the command, and if I download the Standard3dDataset_ V1.2.zip does not require the addition of "-- align_angle".
Can you briefly explain the differences between the new way of data processing and the previous one？
- Only use the non-aligned version dataset and aligned angle by rotation on non-aligned version raw data;
- Render normal vectors as what I introduced in ScanNet.
if ptv2 will perform better when using the new data processing？
Training on S3DIS is not stable, but I observed that the introducing of normal reduces overfitting. (You need to enable normal in the config, refer to the following config file https://github.com/Pointcept/Pointcept/blob/main/configs/s3dis/semseg-spunet-v1m1-0-cn-base.py)
Thank you very much!
May I ask if the Stanford2d3dDataset_noXYZ dataset was downloaded from here?
Yes.