Spaces:
Runtime error
Runtime error
add basic app for pace prediction
Browse files
app.py
ADDED
@@ -0,0 +1,83 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from pathlib import Path
|
2 |
+
|
3 |
+
import numpy as np
|
4 |
+
import tensorflow as tf
|
5 |
+
import gradio as gr
|
6 |
+
|
7 |
+
import cv2
|
8 |
+
import keras
|
9 |
+
from keras import Sequential
|
10 |
+
from keras.layers import Flatten, Dense
|
11 |
+
|
12 |
+
model_weights_path = (Path.cwd() / "pace_model_weights.h5").resolve()
|
13 |
+
height, width, channels = (224, 224, 3)
|
14 |
+
|
15 |
+
class PaceModel:
|
16 |
+
def __init__(self, height, width, channels):
|
17 |
+
self.resnet_model = Sequential()
|
18 |
+
self.height = height
|
19 |
+
self.width = width
|
20 |
+
self.channels = channels
|
21 |
+
self.class_names = ["Fast", "Medium", "Slow"]
|
22 |
+
|
23 |
+
self.create_pretrained()
|
24 |
+
self.create_architecture()
|
25 |
+
|
26 |
+
def create_pretrained(self):
|
27 |
+
self.pretrained_model = tf.keras.applications.ResNet50(
|
28 |
+
include_top=False,
|
29 |
+
input_shape=(self.height, self.width, self.channels),
|
30 |
+
pooling="avg",
|
31 |
+
classes=211,
|
32 |
+
weights="imagenet"
|
33 |
+
)
|
34 |
+
|
35 |
+
for layer in self.pretrained_model.layers:
|
36 |
+
layer.trainable = False
|
37 |
+
|
38 |
+
def create_architecture(self):
|
39 |
+
self.resnet_model.add(self.pretrained_model)
|
40 |
+
self.resnet_model.add(Flatten())
|
41 |
+
self.resnet_model.add(Dense(1024, activation="relu"))
|
42 |
+
self.resnet_model.add(Dense(256, activation="relu"))
|
43 |
+
self.resnet_model.add(Dense(3, activation="softmax"))
|
44 |
+
|
45 |
+
self.resnet_model.load_weights(model_weights_path)
|
46 |
+
|
47 |
+
def predict(self, input_image: np.ndarray):
|
48 |
+
resized_image = cv2.resize(input_image, (self.height, self.width))
|
49 |
+
image = np.expand_dims(resized_image, axis=0)
|
50 |
+
|
51 |
+
prediction = self.resnet_model.predict(image)
|
52 |
+
print(prediction, np.argmax(prediction))
|
53 |
+
return self.class_names[np.argmax(prediction)]
|
54 |
+
|
55 |
+
def main():
|
56 |
+
model = PaceModel(height, width, channels)
|
57 |
+
|
58 |
+
demo = gr.Interface(
|
59 |
+
fn=model.predict,
|
60 |
+
inputs=gr.Image(
|
61 |
+
type="numpy",
|
62 |
+
label="Upload an image",
|
63 |
+
show_label=True,
|
64 |
+
container=True
|
65 |
+
),
|
66 |
+
outputs=gr.Textbox(
|
67 |
+
lines=1,
|
68 |
+
placeholder="Fast | Medium | Slow",
|
69 |
+
label="Pace of the image",
|
70 |
+
show_label=True,
|
71 |
+
container=True,
|
72 |
+
type="text"
|
73 |
+
),
|
74 |
+
cache_examples=False,
|
75 |
+
live=False,
|
76 |
+
title="Predict Pace",
|
77 |
+
description="Provide an image to determine the pace of the image",
|
78 |
+
)
|
79 |
+
|
80 |
+
demo.queue().launch()
|
81 |
+
|
82 |
+
if __name__ == "__main__":
|
83 |
+
main()
|