diff --git a/.gitignore b/.gitignore
index d4150f0..fb7775f 100644
--- a/.gitignore
+++ b/.gitignore
@@ -1,2 +1,5 @@
 .vs/
 *.onnx
+venv/
+.vscode/
+*.jpg
diff --git a/py/segm.py b/py/segm.py
new file mode 100644
index 0000000..a3d9af9
--- /dev/null
+++ b/py/segm.py
@@ -0,0 +1,45 @@
+import glob
+import os
+
+import cv2
+import numpy as np
+import onnxruntime
+
+
+def main():
+    session = onnxruntime.InferenceSession(
+        "../TongueSegmentation/ResUNet.onnx",
+        providers=["CPUExecutionProvider"],
+    )
+
+    for input_ in session.get_inputs():
+        print(input_.name, input_.shape, input_.type)
+
+    for output in session.get_outputs():
+        print(output.name, output.shape, output.type)
+
+    image_paths = glob.glob(os.path.join(".", "*.jpg"))
+
+    for image_path in image_paths:
+        x = cv2.imread(image_path)
+        x = cv2.resize(x, (256, 256))
+        cv2.imshow("input", x)
+        x = x.astype(np.float32) / 255.0
+        x = np.array([x.transpose(2, 0, 1)])  # HWC -> CHW
+        print(x.shape)
+
+        input_name = session.get_inputs()[0].name
+        output_names = [output.name for output in session.get_outputs()]
+        output = session.run(output_names, {input_name: x})
+        output_array = np.array(output[0][0][0])
+
+        probability_map = 1.0 / (1.0 + np.power(np.e, -output_array))
+        _, tongue_mask = cv2.threshold(probability_map, 0.5, 1, cv2.THRESH_BINARY)
+        tongue_mask = tongue_mask.astype(np.uint8)
+
+        cv2.imshow("tongue_mask", tongue_mask * 255)
+        cv2.waitKey(0)
+
+
+if __name__ == "__main__":
+    main()