geo-deep-learning icon indicating copy to clipboard operation
geo-deep-learning copied to clipboard

Published 20 hours ago • verified publisher icon NRCan

problems with VRT Dataset Output

Open valhassan opened this issue 2 years ago • 3 comments

Multiband VRT data is created and returned as a raster input (String XML) this is a departure from the previous functionality in which a raster file is opened as a rasterio object with properties such as (.name, .count, .meta). We are unable to use similar properties from a VRT dataset hence the output is misinterpreted and gives rise to a variety of problems. VRT_name_error_1 VRT_name_error

valhassan avatar Oct 07 '22 17:10 valhassan

Can you detail the problems you are facing? The naming of output patches is very ugly, I agree, (</VRTDataset>...tif) but hasn't been too problematic to me for now. Are you facing other problems?

FYI, I tried to play with the softcoding the output name of patches but if I recall correctly it implies some minor changes to solaris which I was hoping not to get into...

remtav avatar Oct 12 '22 14:10 remtav

Writing to file inference images using the .name property does not work with VRT Dataset. Using other properties incorrectly with the VRT dataset might be propagating errors at inference.

valhassan avatar Oct 12 '22 14:10 valhassan

For reference, debugging from these lines will help document this issue: https://github.com/NRCan/geo-deep-learning/blob/develop/dataset/aoi.py#L214 https://github.com/NRCan/geo-deep-learning/blob/develop/inference_segmentation.py#L412 Could be reproduced in theory by running an inference from test data.

remtav avatar Oct 13 '22 17:10 remtav