add finetuning config with helios for forest loss driver classification

.github/workflows/forest_loss_driver_prediction.yaml

@@ -89,8 +89,8 @@ jobs:
         env:
           EXTRA_ARGS: |
             [
-              "--pred_pipeline_config",
-              "rslp/forest_loss_driver/inference/config/forest_loss_driver_predict_pipeline_config.yaml",
+              "--integrated_config",
+              "rslp/forest_loss_driver/config/forest_loss_driver_predict_pipeline_config.yaml",
             ]
           CLUSTERS: |
             ["ai2/jupiter-cirrascale-2"]

data/forest_loss_driver/README.md

@@ -0,0 +1,23 @@
+Dataset Versions
+----------------
+
+- 20240912: original version of dataset.
+- 20250424: fix point labels so that they are contained within the window. This is both
+  for compatibility with ES Studio and because new version of rslearn VectorFormat only
+  loads features within the window bounds.
+- 20250429: remove Planet images and replace Sentinel-2 L1C images with L2A images from
+  Planetary Computer which are stored as GeoTIFFs.
+- 20250514: like 20250424 (with Sentinel-2 L1C images + Planet images) but put the
+  polygon in the GeoJSON instead of the point (so when we import to ES Studio it shows
+  up nicely with the forest loss polygon) and update items.json to include the best_X
+  layers (so that the timestamps appear for those layers in ES Studio).
+- 20250605: keep the Planet images but get 6 pre and 6 post Sentinel-2 L2A, Sentinel-1,
+  and Landsat images from Planetary Computer and AWS.
+
+
+Dataset Configurations
+----------------------
+
+- config.json: current inference config that uses RGB PNGs.
+- config_ms.json: corresponds to dataset 20250429, it gets L2A images with all bands
+  stored as GeoTIFF.