search

nasaharvest / crop-mask

End-to-end workflow for generating high resolution cropland maps
Apache License 2.0
95 stars 28 forks source link

Add mali sikasso labels #39

Closed ivanzvonkov closed 3 years ago

ivanzvonkov commented 3 years ago

Not exported on Google Earth yet because of a bit of a backlog there.

hannah-rae commented 3 years ago

No, we know that it is cropland but we don’t know which crop it is (usually because there was an error filling out the form).

On Mon, May 3, 2021 at 6:48 AM Ivan Zvonkov @.***> wrote:

@.**** commented on this pull request.

In data/datasets_labeled.py https://github.com/nasaharvest/crop-mask/pull/39#discussion_r625097625:

@@ -162,6 +162,12 @@ def clean_geowiki(df: pd.DataFrame) -> pd.DataFrame: end_year=2020, train_val_test=(0.8, 0.1, 0.1), ),

  • Processor(
  • filename="sikasso_clean_fields",
  • crop_prob=1.0,

Am I correctly understanding that Crop type is None means we don't know if it is or is not cropland? If the above is true, then the crop_probability=0.5 and these labels will be disregarded when creating the features.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/nasaharvest/crop-mask/pull/39#discussion_r625097625, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABQPKBB7GY2OWS4EXPPNAQTTL2SRNANCNFSM435VNFCQ .

-- Hannah Kerner, PhD Assistant Research Professor University of Maryland/NASA Harvest http://www.public.asu.edu/~hkernerhttps://hannah-rae.github.io/

ivanzvonkov commented 3 years ago

@hannah-rae Okay then the model metrics quoted here are correct.

FYI we have a lot more crop than non-crop for Mali image

hannah-rae commented 3 years ago

FYI we have a lot more crop than non-crop for Mali

Yes, this is expected because the crop labels are from the field data collection efforts but the non-crop labels were hand drawn later by me.