Initial Experiment for Mozambique LULC #207
Conversation
| init_args: | ||
| size: 4 | ||
| mode: "center" | ||
| image_selectors: ["sentinel2_l2a", "sentinel1", "target/crop_type_classification/classes", "target/crop_type_classification/valid"] |
There was a problem hiding this comment.
It makes sense that we need to pad (or crop) both the inputs and the targets - the naming convention for the targets here is confusing to me.
| init_args: | ||
| band_names: | ||
| sentinel2_l2a: ["B02", "B03", "B04", "B08", "B05", "B06", "B07", "B8A", "B11", "B12", "B01", "B09"] | ||
| train_config: |
There was a problem hiding this comment.
I don't think these keys are necessary for ES run? Could we get rid of everything under "task"?
| patch_size: 1 | ||
| decoders: | ||
| crop_type_classification: | ||
| - class_path: rslp.crop.kenya_nandi.train.SegmentationPoolingDecoder |
There was a problem hiding this comment.
@gabrieltseng just a heads-up: I merged the Nandi esrun PR and reorganized the folders, it’s better not to keep all Ag/LULC projects under crop folder. You’ll need to update your import here to rslp.nandi.train.SegmentationPoolingDecoder instead of rslp.crop.kenya_nandi.train.SegmentationPoolingDecoder, and you can move the mozambique folder directly under esrun_data and rslp to align with the data/helios structure. Let me know if you hit any issues and I can help fix.
| "max_matches": 12 | ||
| }, | ||
| "sort_by": "eo:cloud_cover", | ||
| "time_offset": "-180d" |
There was a problem hiding this comment.
Growing season is October - July. This is the date range for which the labels are relevant.
Create datasets for three provinces in Mozambique, Initial runs using S1 + S2