guochengqian / PointNeXt

[NeurIPS'22] PointNeXt: Revisiting PointNet++ with Improved Training and Scaling Strategies
https://guochengqian.github.io/PointNeXt/
MIT License
760 stars 108 forks source link

How to visualize Scannet dataset? #144

Open Qiyuan-Liu opened 8 months ago

Qiyuan-Liu commented 8 months ago

Hi PointNeXt team,

Is there a way to generate the predicted ScanNet point cloud and visualize them?

Looking forward to your reply. Hope you have a great one.

Best regards,

Liu

HJ-Xu commented 7 months ago

Hi Liu

Has your issue been resolved? I would like to know how to visualize the Shapenetpart and S3DIS datasets. Do you have any suggestions on how to implement this?

Looking forward to your reply.

Best regards,

Xu

Ahren606 commented 7 months ago

You need to create a cmap for color mapping in /openpoints/dataset/scannetv2.py by following the example of the s3dis file. Then, when running the test command, add visualize=True. Since the test data does not have labels, only generate .obj files for the input and the PointNeXt-XL prediction results. If ground truth visualization is needed, replace the data in the test folder with the data from the val folder and change no_label=True to False in the test command line,And change None to data[2] inhttps://github.com/guochengqian/PointNeXt/blob/8a0ec91de5329b3b3e346422fa43a51ba24fd450/examples/segmentation/main.py#L76.