Spaces:
Sleeping
Sleeping
Upload functions.py
Browse files- functions.py +47 -0
functions.py
ADDED
@@ -0,0 +1,47 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import numpy as np
|
2 |
+
import json
|
3 |
+
import onnxruntime as rt
|
4 |
+
|
5 |
+
model_path = 'model/model.onnx'
|
6 |
+
class_path = 'model/birds_name_mapping.json'
|
7 |
+
|
8 |
+
normalise_means = [0.4914, 0.4822, 0.4465]
|
9 |
+
normalise_stds = [0.2023, 0.1994, 0.2010]
|
10 |
+
|
11 |
+
def normalise_image(image):
|
12 |
+
image = image.copy()
|
13 |
+
for i in range(3):
|
14 |
+
image[:, i, :, :] = (image[:, i, :, :] - normalise_means[i]) / normalise_stds[i]
|
15 |
+
return image
|
16 |
+
|
17 |
+
def load_class_names():
|
18 |
+
with open(class_path, 'r') as f:
|
19 |
+
class_names = json.load(f)
|
20 |
+
return class_names
|
21 |
+
|
22 |
+
def predict(inp_image):
|
23 |
+
|
24 |
+
class_names = load_class_names()
|
25 |
+
|
26 |
+
image = inp_image
|
27 |
+
image = image.transpose((2, 0, 1))
|
28 |
+
|
29 |
+
image = image / 255.0
|
30 |
+
image = np.expand_dims(image, axis=0)
|
31 |
+
image = normalise_image(image)
|
32 |
+
image = image.astype(np.float32)
|
33 |
+
|
34 |
+
sess = rt.InferenceSession(model_path)
|
35 |
+
|
36 |
+
input_name = sess.get_inputs()[0].name
|
37 |
+
output_name = sess.get_outputs()[0].name
|
38 |
+
|
39 |
+
output = sess.run([output_name], {input_name: image})[0]
|
40 |
+
prob = np.exp(output) / np.sum(np.exp(output), axis=1, keepdims=True)
|
41 |
+
|
42 |
+
top5 = np.argsort(prob[0])[-5:][::-1]
|
43 |
+
|
44 |
+
class_probs = {class_names[str(i)]: float(prob[0][i]) for i in top5}
|
45 |
+
print(class_probs)
|
46 |
+
|
47 |
+
return class_probs
|