drkareemkamal
commited on
Commit
•
5333123
1
Parent(s):
a3fbc07
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,88 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
|
3 |
+
from timeit import default_timer as timer
|
4 |
+
from typing import Tuple , Dict
|
5 |
+
import tensorflow as tf
|
6 |
+
import numpy as np
|
7 |
+
|
8 |
+
from PIL import Image
|
9 |
+
import os
|
10 |
+
|
11 |
+
# 1.Import and class names setup
|
12 |
+
class_names = ['CNV','DME','DRUSEN','NORMAL']
|
13 |
+
|
14 |
+
|
15 |
+
# 2. Model annd transforms prepration
|
16 |
+
# model = tf.keras.models.load_model(
|
17 |
+
# 'oct_classification_final_model_lg.keras', custom_objects=None, compile=True, safe_mode=True
|
18 |
+
# )
|
19 |
+
model = tf.keras.models.load_model(
|
20 |
+
'oct_classification_final_model_lg.keras', custom_objects=None, compile=True, safe_mode=False
|
21 |
+
)
|
22 |
+
|
23 |
+
|
24 |
+
# Load save weights
|
25 |
+
|
26 |
+
# 3.prediction function (predict())
|
27 |
+
|
28 |
+
def load_and_prep_imgg(img : Image.Image, img_shape=224, scale=True):
|
29 |
+
# if not isinstance(filename, str):
|
30 |
+
# raise ValueError("The filename must be a string representing the file path.")
|
31 |
+
# img = tf.io.read_file(filename)
|
32 |
+
# img = tf.io.decode_image(img, channels=3)
|
33 |
+
# img = tf.image.resize(img, size=[img_shape, img_shape])
|
34 |
+
# if scale:
|
35 |
+
# return img / 255
|
36 |
+
# else:
|
37 |
+
# return img
|
38 |
+
img = img.resize((img_shape, img_shape))
|
39 |
+
img = np.array(img)
|
40 |
+
if img.shape[-1] == 1: # If the image is grayscale
|
41 |
+
img = np.stack([img] * 3, axis=-1)
|
42 |
+
img = tf.convert_to_tensor(img, dtype=tf.float32)
|
43 |
+
if scale:
|
44 |
+
return img / 255.0
|
45 |
+
else:
|
46 |
+
return img
|
47 |
+
|
48 |
+
def predict(img) -> Tuple[Dict,float] :
|
49 |
+
|
50 |
+
start_time = timer()
|
51 |
+
|
52 |
+
image = load_and_prep_imgg(img)
|
53 |
+
#image = Image.open(image)
|
54 |
+
|
55 |
+
pred_img = model.predict(tf.expand_dims(image, axis=0))
|
56 |
+
pred_class = class_names[pred_img.argmax()]
|
57 |
+
print(f"Predicted macular diseases is: {pred_class} with probability: {pred_img.max():.2f}")
|
58 |
+
|
59 |
+
|
60 |
+
|
61 |
+
|
62 |
+
end_time = timer()
|
63 |
+
pred_time = round(end_time - start_time , 4)
|
64 |
+
|
65 |
+
return pred_class , pred_time
|
66 |
+
|
67 |
+
### 4. Gradio app - our Gradio interface + launch command
|
68 |
+
|
69 |
+
title = 'Macular Disease Classification'
|
70 |
+
description = 'Feature Extraction VGG model to classify Macular Diseases by OCT'
|
71 |
+
article = 'Created with TensorFlow Model Deployment'
|
72 |
+
# Create example list
|
73 |
+
|
74 |
+
example_list = [['examples/'+ example] for example in os.listdir('examples')]
|
75 |
+
example_list
|
76 |
+
|
77 |
+
# create a gradio demo
|
78 |
+
demo = gr.Interface(fn=predict ,
|
79 |
+
inputs=gr.Image(type='pil'),
|
80 |
+
outputs=[gr.Label(num_top_classes = 3 , label= 'prediction'),
|
81 |
+
gr.Number(label= 'Prediction time (s)')],
|
82 |
+
examples = example_list,
|
83 |
+
title = title,
|
84 |
+
description = description,
|
85 |
+
article= article)
|
86 |
+
|
87 |
+
# Launch the demo
|
88 |
+
demo.launch(debug= False)
|