search

azavea / raster-vision

An open source library and framework for deep learning on satellite and aerial imagery.
https://docs.rastervision.io
Other
2.08k stars 388 forks source link

Avoid making training chips on NODATA chips for semseg #718

Closed yoninachmany closed 2 years ago

yoninachmany commented 5 years ago

Like https://github.com/azavea/raster-vision/pull/364, check chip before including its resulting data.

lewfish commented 5 years ago

We do this for object detection here and chip classification here. It appears that we are not implementing this for semantic segmentation.

As a workaround, you could add an AOI to the scene that avoids the NODATA areas.

AdeelH commented 2 years ago

1025