Add demo(Python ONNX): 310_attentive-gan-derainnet

310_attentive-gan-derainnet/README.md

@@ -3,3 +3,20 @@
 ![image](https://user-images.githubusercontent.com/33194443/178708494-f6755de7-7dd9-49e4-9d0c-84d65cc25e1c.png)
 ## After
 ![image](https://user-images.githubusercontent.com/33194443/178708521-5a50c415-2176-440b-b1a4-78b218473623.png)
+
+# Demo projects
+## attentive gan derainnet with ONNX Runtime in Python
+```
+python demo_attentive-gan-derainnet_onnx.py.py
+```
+
+If you want to change the model, specify it with an argument.
+```python
+    parser.add_argument(
+        "--model",
+        type=str,
+        default='attentive_gan_derainnet_240x360/model_float32.onnx',
+    )
+```
+
+

310_attentive-gan-derainnet/demo/demo_attentive-gan-derainnet_onnx.py

@@ -0,0 +1,107 @@
+#!/usr/bin/env python
+# -*- coding: utf-8 -*-
+import copy
+import time
+import argparse
+from typing import Any, Tuple
+
+import cv2
+import numpy as np
+import onnxruntime  # type: ignore
+
+
+def minmax_scale(input_array: np.ndarray) -> np.ndarray:
+    min_val: float = np.min(input_array)
+    max_val: float = np.max(input_array)
+
+    output_array: np.ndarray = (input_array - min_val) * 255.0 / (max_val -
+                                                                  min_val)
+
+    return output_array
+
+
+def run_inference(
+    onnx_session: onnxruntime.InferenceSession,
+    image: np.ndarray,
+) -> np.ndarray:
+    input_detail = onnx_session.get_inputs()[0]
+    input_name: str = input_detail.name
+    input_shape: Tuple[int, int] = input_detail.shape[2:4]
+
+    # Pre process: Resize, Normalize, Transpose, Expand Dims, float32 cast
+    input_image: np.ndarray = cv2.resize(
+        image,
+        dsize=(input_shape[1], input_shape[0]),
+    )
+    input_image = np.divide(np.array(input_image, np.float32), 127.5) - 1.0
+    input_image = input_image.transpose(2, 0, 1)
+    input_image = np.expand_dims(input_image, axis=0)
+    input_image = input_image.astype('float32')
+
+    # Inference
+    result: Any = onnx_session.run(None, {input_name: input_image})
+
+    # Post process: squeeze, MinMax Scale, uint8 cast
+    output_image: np.ndarray = np.squeeze(result[0])
+    for i in range(output_image.shape[2]):
+        output_image[:, :, i] = minmax_scale(output_image[:, :, i])
+    output_image = output_image.astype(np.uint8)
+
+    return output_image
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser()
+
+    parser.add_argument("--image", type=str, default='sample.png')
+    parser.add_argument(
+        "--model",
+        type=str,
+        default='attentive_gan_derainnet_240x360/model_float32.onnx',
+    )
+
+    args = parser.parse_args()
+    model_path: str = args.model
+    image_path: str = args.image
+
+    # Load model
+    onnx_session: onnxruntime.InferenceSession = onnxruntime.InferenceSession(
+        model_path,
+        providers=[
+            'CUDAExecutionProvider',
+            'CPUExecutionProvider',
+        ],
+    )
+
+    image: np.ndarray = cv2.imread(image_path)
+    debug_image: np.ndarray = copy.deepcopy(image)
+    image_height: int = image.shape[0]
+    image_width: int = image.shape[1]
+
+    start_time: float = time.time()
+
+    # Inference execution
+    output_image: np.ndarray = run_inference(
+        onnx_session,
+        image,
+    )
+
+    output_image = cv2.resize(output_image, dsize=(image_width, image_height))
+
+    elapsed_time: float = time.time() - start_time
+
+    # Inference elapsed time
+    cv2.putText(
+        debug_image,
+        "Elapsed Time : " + '{:.1f}'.format(elapsed_time * 1000) + "ms",
+        (10, 30), cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 255, 0), 1, cv2.LINE_AA)
+
+    # Display
+    cv2.imshow('attentive-gan-derainnet Input', debug_image)
+    cv2.imshow('attentive-gan-derainnet Output', output_image)
+    _ = cv2.waitKey(-1)
+    cv2.destroyAllWindows()
+
+
+if __name__ == '__main__':
+    main()