search

Pointcept / PointTransformerV3

[CVPR'24 Oral] Official repository of Point Transformer V3 (PTv3)
MIT License
583 stars 30 forks source link

Custom dataset causes overflow error #41

Open Shubhendu-Jena opened 2 months ago

Shubhendu-Jena commented 2 months ago

Hi, thanks for the great work.

My custom point cloud range is max= [ 39.4270, 41.9689, 153.3401] and min= [-63.4741, -32.5176, 13.7472]. This causes the following line to cause an overflow error :

https://github.com/Pointcept/PointTransformerV3/blob/342656c7956fc1ff59a13e0969bb858003882f49/serialization/hilbert.py#L143

The error is : RuntimeError: self.stride(-1) must be 1 to view Long as Byte (different element sizes), but got 15000

I'd appreciate any help/suggestion on how I can go about resolving this. Should I scale my point clouds?

Thanks in advance

Gofinge commented 2 months ago

Hi, if it is an object-level point cloud, please scale it. If it is a real-world scene, please increase the grid size.

Shubhendu-Jena commented 2 months ago

Hi,

Thank you for the response. I saw your config files where a number of transformations have been applied (including hashing of the grid coordinates). I intend to fine-tune your model on my data (which consists of real-world data of the form [N,3]). Can I do it with just the raw coordinates? Also, since PTV3 pretrained model weights require 6 input channels, how should I go about dealing with that?

Thanks in advance