Attributes
Hello,
I am using the dataset to reproduce the experiments conducted in the paper "CaBuAr: California Burned Areas dataset for delineation", trying to improve the reported results. I am having some difficulties understanding the dataset structure.
In particular, I was looking for the data in the raw/complete folder, but I noticed that there are only 9 files, while in normalized/complete there are 27 files. Why is there this difference?
I also noticed that, for the complete files, the attributes mentioned in the paper (such as "Clouds over the burned area" or "Too many clouds over the image") are not present. How is it possible to filter out low-quality data without these attributes?
Finally, regarding the patched data, the attributes are present. In the paper, you mention that patches are filtered based on these attributes. Is it possible to have the full list of attributes that can be used for filtering?
Thank you in advance for your help.
Best regards,
Samuele Tomassacci
I suggest using the normalized version; they are rescaled. I stopped uploading the raw files because of their size and because they contain the same information as the normalized version (although they can be compressed more effectively). The attributes are the same for both sets, allowing you to map them.
You can find the filter based on the details provided in the paper.