Datasets:
Inquiry regarding missing attributes in the updated DALES-2 dataset
Our team is currently training a semantic segmentation machine learning model using the DALES-2 dataset. In our training pipeline, we rely on the laser reflection properties (specifically return_number and number_of_returns) as well as the geometric attribute scan_angle_rank to ensure high model accuracy.
We have noticed that these attributes are missing in the new, corrected version of the dataset. Could you please clarify if these features were intentionally removed, or if there is a plan to include them in a future update?
Access to these parameters is critical for our model's performance, and your guidance would be greatly appreciated.
Hi,
return_number/number_of_returns and scan_angle_rank were not included in the original DALES benchmark and couldn't be recreated. We kept every property already included in the original benchmark and anything missing would in reality be non-existent.
Those data were included, and below are example visualizations based on the point cloud from tile 5095_54455.
scan_angle_rank:
return_number:
number_of_returns:
We used the DALESObjects data as the basis for our reannotation. I just re-verified my files, and couldn't find these properties. I checked the article as well for any mention of these properties. Pardon my mistake. If you could share the source for your version of DALES, I'll be glad to update the present dataset with these additional properties.
Thank you for the clarification.
To clarify, I am not using the DALESObjects dataset. I am using the following version:
Varney, Nina & Asari, Vijayan & Graehling, Quinn. (2020). DALES: A Large-scale Aerial LiDAR Data Set for Semantic Segmentation. 10.48550/arXiv.2004.11985.
According to the "creation_date" found in the header of these files, this dataset was created on 2019-11-22. This version includes the properties I mentioned.