Accommodate MS and Bands Beyond Default RGB

docs/parameters.rst

@@ -72,3 +72,7 @@ Here is the full list of configuration parameters you can specify in a ``config.
 
 **over_zoom**: int
  	An integer greater than 0. If set for XYZ tiles, it will fetch tiles from `zoom` + `over_zoom`, to create higher resolution tiles which fill out the bounds of the original zoom level.
+
+**band_indices**: list
+	A list of band indices to pull from a tiff. Defaults to (1, 2, 3) if not specified. For using non-RGB tifs, ie MultiSpectral tifs this should be over-written
+	for example to [5, 3, 2, 7] to extract Red, Green, Blue, and NIR bands from the MultiSpectral spacenet challenge imagery.

label_maker/package.py

@@ -4,6 +4,7 @@
 from os import path as op
 from urllib.parse import urlparse
 import numpy as np
+import rasterio
 from PIL import Image
 
 from label_maker.utils import is_tif, get_image_format
@@ -73,25 +74,29 @@ def package_directory(dest_folder, classes, imagery, ml_type, seed=False,
 
     # open the images and load those plus the labels into the final arrays
     if is_tif(imagery):  # if a TIF is provided, use jpg as tile format
-        image_format = '.jpg'
+        img_dtype = rasterio.open(imagery).profile['dtype']
+        image_format = '.tif'
 
     else:
+        img_dtype = np.uint8
         image_format = get_image_format(imagery, kwargs)
 
     for tile in tiles:
         image_file = op.join(dest_folder, 'tiles', '{}{}'.format(tile, image_format))
         try:
-            img = Image.open(image_file)
+            img = rasterio.open(image_file)
         except FileNotFoundError:
             # we often don't download images for each label (e.g. background tiles)
             continue
         except OSError:
             print('Couldn\'t open {}, skipping'.format(image_file))
             continue
 
-        np_image = np.array(img)
+        i = np.array(img.read())
+        np_image = np.moveaxis(i, 0, 2)
         img.close()
 
+
         x_vals.append(np_image)
         if ml_type == 'classification':
             y_vals.append(labels[tile])
@@ -103,7 +108,9 @@ def package_directory(dest_folder, classes, imagery, ml_type, seed=False,
             y_vals.append(labels[tile][..., np.newaxis])  # Add grayscale channel
 
     # Convert lists to numpy arrays
-    x_vals = np.array(x_vals, dtype=np.uint8)
+
+    #TO-DO flexible x_val dtype
+    x_vals = np.array(x_vals, dtype=img_dtype)
     y_vals = np.array(y_vals, dtype=np.uint8)
 
     # Get number of data samples per split from the float proportions
@@ -128,3 +135,4 @@ def package_directory(dest_folder, classes, imagery, ml_type, seed=False,
 
     np.savez(op.join(dest_folder, 'data.npz'), **save_dict)
     print('Saving packaged file to {}'.format(op.join(dest_folder, 'data.npz')))
+    print('Image dtype written in npz matches input image dtype: {}'.format(img_dtype))

label_maker/utils.py

@@ -106,6 +106,7 @@ def get_tile_tif(tile, imagery, folder, kwargs):
     bound = bounds(*[int(t) for t in tile.split('-')])
     imagery_offset = kwargs.get('imagery_offset') or [0, 0]
     with rasterio.open(imagery) as src:
+        profile = src.profile
         x_res, y_res = src.transform[0], src.transform[4]
 
         # offset our imagery in the "destination pixel" space
@@ -137,15 +138,19 @@ def get_tile_tif(tile, imagery, folder, kwargs):
         window = ((top, bottom), (left, right))
 
         # read the first three bands (assumed RGB) of the TIF into an array
-        data = np.empty(shape=(3, 256, 256)).astype(src.profile['dtype'])
-        for k in (1, 2, 3):
-            src.read(k, window=window, out=data[k - 1], boundless=True)
+        band_indices = kwargs.get('band_indices', (1, 2, 3))
+        band_count = len(band_indices)
 
-        # save
-        tile_img = op.join(folder, '{}{}'.format(tile, '.jpg'))
-        img = Image.fromarray(np.moveaxis(data, 0, -1), mode='RGB')
-        img.save(tile_img)
+        arr_shape = (band_count, 256, 256)
+        data = np.empty(shape=(arr_shape)).astype(src.profile['dtype'])
 
+        for i, k in enumerate(band_indices):
+            src.read(k, window=window, out=data[i], boundless=True)
+        # save
+        tile_img = op.join(folder, '{}{}'.format(tile, '.tif'))
+        with rasterio.open(tile_img, 'w', driver='png', height=256,
+                width=256, count=band_count, dtype=profile['dtype']) as w:
+                w.write(data)
     return tile_img
 
 def get_tile_wms(tile, imagery, folder, kwargs):

label_maker/validate.py

@@ -35,5 +35,6 @@
     'split_vals': {'type': 'list', 'schema': {'type': 'float'}},
     'split_names': {'type': 'list', 'schema': {'type': 'string'}},
     'tms_image_format': {'type': 'string'},
-    'over_zoom': {'type': 'integer', 'min': 1}
+    'over_zoom': {'type': 'integer', 'min': 1},
+    'band_indices': {'type': 'list'}
 }

test/fixtures/integration/config.integration.bands.json

@@ -0,0 +1,12 @@
+{
+    "zoom": 18,
+    "classes": [
+        {"name": "paved", "filter": ["in", "paved", "1"]},
+        {"name": "unpaved", "filter": ["in", "paved", "2"]}
+        ],
+    "geojson": "integration-ms/ms-roads.geojson",
+    "imagery": "integration-ms/ms-img.tif",
+    "background_ratio": 0,
+    "ml_type": "classification",
+    "band_indices": [5, 7]
+  }