Merge pull request #9 from earthnet2021/quickfix

fix sen2nbar errors, cloud mask rescaling

README.md

@@ -38,7 +38,7 @@ import earthnet_minicuber as emc
 specs = {
     "lon_lat": (43.598946, 3.087414), # center pixel
     "xy_shape": (256, 256), # width, height of cutout around center pixel
-    "resolution": 20, # in meters.. will use this on a local UTM grid..
+    "resolution": 10, # in meters.. will use this on a local UTM grid..
     "time_interval": "2021-07-01/2021-07-31",
     "providers": [
         {
@@ -94,6 +94,7 @@ Kwargs:
 - `five_daily_filter`: If `True` returns a regular 5-daily cycle starting with the first date in `full_time_interval`. It has no effect, if `best_orbit_filter` is used.
 - `brdf_correction`: If `True`, does BRDF correction based on the Sentinel 2 Metadata (illumination angles).
 - `cloud_mask`: If `True`, creates a cloud and cloud shadow mask based on deep learning. It automatically finds the best available cloud mask for the requested `bands`.
+- `cloud_mask_rescale_factor`: If using cloud mask and a lower resolution than 10m, set this rescaling factor to the multiple of 10m that you are requesting. E.g. if `resolution = 20`, set `cloud_mask_rescale_factor = 2`.
 - `correct_processing_baseline`: If `True` (default): corrects the shift of +1000 that exists in Sentinel 2 data with processing baseline >= 4.0
 
 

earthnet_minicuber/__init__.py

@@ -1,6 +1,6 @@
 """EarthNet Minicuber"""
 
-__version__ = "0.1.2"
+__version__ = "0.1.3"
 __author__ = "Vitus Benson"
 
 

earthnet_minicuber/provider/s2/cloudmask.py

@@ -26,8 +26,9 @@ def get_checkpoint(bands_avail):
 
 
 class CloudMask:
-    def __init__(self, bands = ["B02", "B03", "B04", "B8A"]):
+    def __init__(self, bands = ["B02", "B03", "B04", "B8A"], cloud_mask_rescale_factor = None):
 
+        self.cloud_mask_rescale_factor = cloud_mask_rescale_factor
         self.bands = bands
         ckpt, self.ckpt_bands = get_checkpoint(bands)
 
@@ -63,12 +64,21 @@ def __call__(self, stack):
 
         x = torch.nn.functional.pad(x, (w_pad_left, w_pad_right, h_pad_left, h_pad_right), mode = "reflect")
 
+        if self.cloud_mask_rescale_factor:
+            #orig_size = (x.shape[-2], x.shape[-1])
+            x = torch.nn.functional.interpolate(x, scale_factor = self.cloud_mask_rescale_factor, mode = 'bilinear')
+
         with torch.no_grad():
             y_hat = self.model(x)
 
-        y_hat = y_hat[:, :, h_pad_left:-h_pad_right, w_pad_left:-w_pad_right]
+        y_hat = torch.argmax(y_hat, dim = 1).float()
+
+        if self.cloud_mask_rescale_factor:
+            y_hat = torch.nn.functional.max_pool2d(y_hat[:,None,...], kernel_size = self.cloud_mask_rescale_factor)[:,0,...]#torch.nn.functional.interpolate(y_hat, size = orig_size, mode = "bilinear")
+
+        y_hat = y_hat[:, h_pad_left:-h_pad_right, w_pad_left:-w_pad_right]
 
-        ds["mask"] = (("time", "y", "x"),torch.argmax(y_hat, dim = 1).float().cpu().numpy())
+        ds["mask"] = (("time", "y", "x"),y_hat.cpu().numpy())
 
         return ds.to_array("band")
 

earthnet_minicuber/provider/s2/nbar.py

@@ -53,13 +53,16 @@ def call_sen2nbar(stack, items, epsg):
     # Compute the c-factor per item and extract the processing baseline
     c_array = []
     for item in ordered_items:#tqdm(ordered_items, disable=quiet):
-        c = c_factor_from_item(item, f"epsg:{epsg}")
-        c = c.interp(
-            y=stack.y.values,
-            x=stack.x.values,
-            method="linear",
-            kwargs={"fill_value": "extrapolate"},
-        )
+        try:
+            c = c_factor_from_item(item, f"epsg:{epsg}")
+            c = c.interp(
+                y=stack.y.values,
+                x=stack.x.values,
+                method="linear",
+                kwargs={"fill_value": "extrapolate"},
+            )
+        except ValueError:
+            c = xr.DataArray(np.full((9,len(stack.y), len(stack.x)), np.NaN), coords = {"band": ['B02','B03','B04','B05','B06','B07','B08','B11','B12'], "y": stack.y, "x": stack.x}, dims = ("band", "y", "x"))
         c_array.append(c)
 
     orig_bands = stack.band.values.tolist()

earthnet_minicuber/provider/s2/sentinel2.py

@@ -39,11 +39,11 @@
 
 class Sentinel2(provider_base.Provider):
 
-    def __init__(self, bands = ["AOT", "B01", "B02", "B03", "B04", "B05", "B06", "B07", "B08", "B8A", "B09", "B11", "B12", "WVP"], best_orbit_filter = True, five_daily_filter = False, brdf_correction = True, cloud_mask = True, aws_bucket = "planetary_computer", s2_avail_var = True, correct_processing_baseline = True):
+    def __init__(self, bands = ["AOT", "B01", "B02", "B03", "B04", "B05", "B06", "B07", "B08", "B8A", "B09", "B11", "B12", "WVP"], best_orbit_filter = True, five_daily_filter = False, brdf_correction = True, cloud_mask = True, cloud_mask_rescale_factor = None, aws_bucket = "planetary_computer", s2_avail_var = True, correct_processing_baseline = True):
 
         self.is_temporal = True
 
-        self.cloud_mask = CloudMask(bands=bands) if cloud_mask else None
+        self.cloud_mask = CloudMask(bands=bands, cloud_mask_rescale_factor = cloud_mask_rescale_factor) if cloud_mask else None
 
         if self.cloud_mask and "SCL" not in bands:
             bands += ["SCL"]
@@ -201,6 +201,9 @@ def load_data(self, bbox, time_interval, **kwargs):
 
                 stack = stack.sel(time = stack.time.dt.date.isin(dates))
 
+            if len(stack.time) == 0:
+                return None
+
             if self.correct_processing_baseline:
                 stack = correct_processing_baseline(stack, items_s2)
 

setup.py

@@ -31,7 +31,7 @@
 
 
 setup(name='earthnet-minicuber', 
-        version='0.1.2',
+        version='0.1.3',
         description="EarthNet Minicuber",
         author="Vitus Benson, Christian Requena-Mesa",
         author_email="[email protected]",