Upload app.py
Browse files
app.py
ADDED
@@ -0,0 +1,65 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import tensorflow as tf
|
3 |
+
import numpy as np
|
4 |
+
from PIL import Image
|
5 |
+
|
6 |
+
# Make sure to decorate the custom metric class with @register_keras_serializable
|
7 |
+
@tf.keras.utils.register_keras_serializable()
|
8 |
+
class F1Score(tf.keras.metrics.Metric):
|
9 |
+
def __init__(self, name='f1_score', **kwargs):
|
10 |
+
super().__init__(name=name, **kwargs)
|
11 |
+
self.precision = tf.keras.metrics.Precision()
|
12 |
+
self.recall = tf.keras.metrics.Recall()
|
13 |
+
|
14 |
+
def update_state(self, y_true, y_pred, sample_weight=None):
|
15 |
+
y_pred = tf.round(y_pred)
|
16 |
+
self.precision.update_state(y_true, y_pred, sample_weight)
|
17 |
+
self.recall.update_state(y_true, y_pred, sample_weight)
|
18 |
+
|
19 |
+
def result(self):
|
20 |
+
p = self.precision.result()
|
21 |
+
r = self.recall.result()
|
22 |
+
return 2 * ((p * r) / (p + r + tf.keras.backend.epsilon()))
|
23 |
+
|
24 |
+
def reset_states(self):
|
25 |
+
self.precision.reset_states()
|
26 |
+
self.recall.reset_states()
|
27 |
+
|
28 |
+
# Load your TensorFlow model
|
29 |
+
model_path = './ACC0.9322_valACC0.9148_loss0.3592_Epoch83.keras'
|
30 |
+
model = tf.keras.models.load_model(model_path, custom_objects={'F1Score': F1Score})
|
31 |
+
|
32 |
+
def predict(image):
|
33 |
+
# Preprocess the image to match the input shape of the model
|
34 |
+
img = Image.fromarray(image.astype('uint8'), 'RGB')
|
35 |
+
img = img.resize((224, 224)) # Update the size if different
|
36 |
+
img_array = np.array(img)
|
37 |
+
img_array = img_array / 255.0 # Normalize if your model expects normalization
|
38 |
+
img_array = img_array[np.newaxis, ...] # Model expects a batch dimension
|
39 |
+
|
40 |
+
# Make prediction
|
41 |
+
prediction = model.predict(img_array)
|
42 |
+
# Calculate prediction probability
|
43 |
+
prediction_probability = float(prediction[0][0])
|
44 |
+
|
45 |
+
# Check if prediction is 'Normal' or 'Pneumonia'
|
46 |
+
if prediction_probability <= 0.5:
|
47 |
+
# If 'Normal', add 70% to its probability but cap it at 1.0 (100%)
|
48 |
+
adjusted_probability = min(prediction_probability + 0.5, 1.0)
|
49 |
+
pred_class = 'Normal'
|
50 |
+
else:
|
51 |
+
# If 'Pneumonia', use the original probability
|
52 |
+
adjusted_probability = prediction_probability
|
53 |
+
pred_class = 'Pneumonia'
|
54 |
+
|
55 |
+
# Return the prediction class and the (adjusted) probability
|
56 |
+
return {pred_class: adjusted_probability}
|
57 |
+
|
58 |
+
iface = gr.Interface(fn=predict,
|
59 |
+
inputs=gr.components.Image(image_mode='RGB', label="Upload X-ray Image"),
|
60 |
+
outputs=gr.components.Label(num_top_classes=2),
|
61 |
+
title="Pneumonia Detection",
|
62 |
+
description="Upload an X-ray image to detect pneumonia.")
|
63 |
+
|
64 |
+
if __name__ == "__main__":
|
65 |
+
iface.launch(share=True)
|