Merge pull request #9 from dlr-eoc/update_onnxruntime

Update onnxruntime

CHANGELOG.rst

@@ -1,6 +1,12 @@
 Changelog
 =========
 
+[0.1.6] (2022-11-30)
+--------------------
+Changed
+*******
+- update onnxruntime version and add providers to session
+
 [0.1.5] (2022-11-11)
 --------------------
 Changed

README.md

@@ -71,7 +71,7 @@ The easiest way to install ukis-csmask is through pip. The default installation
 pip install ukis-csmask
 ```
 
-To install ukis-csmask with GPU support run the following instead. This requires that you have a GPU with CUDA runtime libraries (CUDA 10.2 and cuDNN 8.0.3) installed on the system.
+To install ukis-csmask with GPU support run the following instead. This requires that you have a GPU with CUDA runtime libraries installed on the system.
 
 ```shell
 pip install ukis-csmask[gpu]

requirements.txt

@@ -1,3 +1,3 @@
 numpy
-onnxruntime
+onnxruntime>=1.13
 scipy

tests/test_mask.py

@@ -70,8 +70,8 @@ def test_csmask_valid(data):
     csmask = CSmask(img=data["img"], band_order=["Blue", "Green", "Red", "NIR", "SWIR1", "SWIR2"])
     y_pred = csmask.valid
     y_true = reclassify(data["msk"], {"reclass_value_from": [0, 1, 2, 3, 4], "reclass_value_to": [0, 1, 1, 1, 0]})
-    y_true_inverted = ~y_true.astype(np.bool)
-    y_true = (~ndimage.binary_dilation(y_true_inverted, iterations=4).astype(np.bool)).astype(np.uint8)
+    y_true_inverted = ~y_true.astype(bool)
+    y_true = (~ndimage.binary_dilation(y_true_inverted, iterations=4).astype(bool)).astype(np.uint8)
     y_true = y_true.ravel()
     y_pred = y_pred.ravel()
     kappa = round(cohen_kappa_score(y_true, y_pred), 2)

ukis_csmask/__init__.py

@@ -1 +1 @@
-__version__ = "0.1.5"
+__version__ = "0.1.6"

ukis_csmask/mask.py

@@ -82,7 +82,9 @@ def _csm(self):
         x /= [0.16431, 0.16762, 0.18230, 0.17409, 0.16020, 0.14164]
 
         # start onnx inference session and load model
-        sess = onnxruntime.InferenceSession(str(Path(__file__).parent) + "/model.onnx")
+        sess = onnxruntime.InferenceSession(
+            str(Path(__file__).parent) + "/model.onnx", providers=onnxruntime.get_available_providers()
+        )
 
         # predict on array tiles
         y_prob = [sess.run(None, {"input_1": tile[np.newaxis, :]}) for n, tile in enumerate(list(x))]
@@ -116,8 +118,8 @@ def _valid(self, invalid_buffer):
 
         # dilate the inverse of the binary valid pixel mask (invalid=0)
         # this effectively buffers the invalid pixels
-        valid_i = ~valid.astype(np.bool)
-        valid = (~ndimage.binary_dilation(valid_i, iterations=invalid_buffer).astype(np.bool)).astype(np.uint8)
+        valid_i = ~valid.astype(bool)
+        valid = (~ndimage.binary_dilation(valid_i, iterations=invalid_buffer).astype(bool)).astype(np.uint8)
 
         if self.nodata_value is not None:
             # add image nodata pixels to valid pixel mask