search

allenai / satlas

Apache License 2.0
202 stars 24 forks source link

Clarification on Matching NAIP Images with Labels in Dataset #50

Closed Amazingren closed 3 weeks ago

Amazingren commented 3 weeks ago

Dear team,

Thank you for sharing such an interesting dataset and work.

While exploring the dataset, I am focusing on using only the NAIP images and their corresponding labels. Specifically, I plan to work with the NAIP data from 2020 and its associated labels.

I noticed that the NAIP images are named in the format xxxxx_xxxxx.png, where x represents a number. However, when I downloaded the static labels (approximately 117GB), they were organized into folders, each named using the format xxxx_xxxx, without matching the exact file names of the NAIP images.

This lack of direct pairing between NAIP images and their labels—unlike what I observed with other data types like Sentinel-1 and Sentinel-2—has caused some confusion in aligning the data with the labels I want to use.

Could you please provide guidance on how to accurately match the labels with the NAIP images?

Thank you very much for your assistance.

Best regards,

favyen2 commented 3 weeks ago

NAIP images are stored at zoom 17 instead of 13 (factor of 16 higher resolution) so 12345_67890.png would correspond to the crop from (288, 64) to (320, 96) in 771_4243.json (divide by 16 and round down to get the zoom 13 tile, and take modulo to get the index within the tile).

Amazingren commented 3 weeks ago

NAIP images are stored at zoom 17 instead of 13 (factor of 16 higher resolution) so 12345_67890.png would correspond to the crop from (288, 64) to (320, 96) in 771_4243.json (divide by 16 and round down to get the zoom 13 tile, and take modulo to get the index within the tile).

Great! Thanks for your quick reply. Now I am clear about this.

Best regards and many thanks,