To many patches after tiling (for flood modelling interest)

NRCan / geo-deep-learning

Deep learning applied to georeferenced datasets

https://geo-deep-learning.readthedocs.io/en/latest/

MIT License

149 stars 49 forks source link

To many patches after tiling (for flood modelling interest) #499

Open Abdielfer opened 1 year ago

Abdielfer commented 1 year ago

After tiling, the destination folder contains all possible patches from the raster image, event when there is not ground truth (GT) labels matching (< min_annot_perc). Flood happens close to water bodies, and in general, flood extensions represents a small percent of surface from the original images. Therefore, majority of patch have no GT match. Should a cleaning process be implemented into DGL (automatically or not) or as an independent script?

mpelchat04 commented 1 year ago

Do you mean that the 'min_annot_perc' does not filter data with your dataset? I'm not sure I understand the issue here.