Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -7,240 +7,172 @@ import tensorflow_hub as hub
|
|
7 |
import numpy as np
|
8 |
from PIL import Image
|
9 |
import io
|
10 |
-
|
11 |
|
12 |
-
# Set up logging
|
13 |
logging.basicConfig(
|
14 |
level=logging.INFO,
|
15 |
format='%(asctime)s - %(name)s - %(levelname)s - %(message)s'
|
16 |
)
|
17 |
logger = logging.getLogger(__name__)
|
18 |
|
19 |
-
|
20 |
-
|
21 |
-
|
22 |
-
|
|
|
|
|
|
|
23 |
|
24 |
-
|
25 |
-
|
26 |
-
|
27 |
-
if not os.path.exists(self.model_path):
|
28 |
-
raise FileNotFoundError(f"Model file not found at {self.model_path}")
|
29 |
-
|
30 |
-
logger.info(f"Loading model from {self.model_path}")
|
31 |
-
|
32 |
-
# Define custom objects dictionary for transfer learning
|
33 |
-
custom_objects = {
|
34 |
-
'KerasLayer': hub.KerasLayer
|
35 |
-
}
|
36 |
-
|
37 |
-
try:
|
38 |
-
logger.info("Attempting to load model with custom objects...")
|
39 |
-
with tf_keras.utils.custom_object_scope(custom_objects):
|
40 |
-
model = tf_keras.models.load_model(self.model_path, compile=False)
|
41 |
-
except Exception as e:
|
42 |
-
logger.error(f"Failed to load with custom objects: {str(e)}")
|
43 |
-
logger.info("Attempting to load model without custom objects...")
|
44 |
-
model = tf_keras.models.load_model(self.model_path, compile=False)
|
45 |
-
|
46 |
-
model.summary()
|
47 |
-
logger.info("Model loaded successfully")
|
48 |
-
return model
|
49 |
-
|
50 |
-
except Exception as e:
|
51 |
-
logger.error(f"Error loading model: {str(e)}")
|
52 |
-
return None
|
53 |
-
|
54 |
-
def preprocess_image(self, image: np.ndarray) -> np.ndarray:
|
55 |
-
"""Preprocess the input image for model prediction."""
|
56 |
-
try:
|
57 |
-
logger.info(f"Input image shape: {image.shape}, dtype: {image.dtype}")
|
58 |
-
|
59 |
-
# If image is RGBA, convert to RGB
|
60 |
-
if image.shape[-1] == 4:
|
61 |
-
logger.info("Converting RGBA to RGB")
|
62 |
-
# Convert to PIL Image and back to handle RGBA->RGB conversion
|
63 |
-
image = Image.fromarray(image).convert('RGB')
|
64 |
-
image = np.array(image)
|
65 |
-
|
66 |
-
# Resize image
|
67 |
-
image = tf_keras.preprocessing.image.smart_resize(
|
68 |
-
image, (256, 256), interpolation='bilinear'
|
69 |
-
)
|
70 |
-
|
71 |
-
# Ensure values are between 0 and 1
|
72 |
-
if image.max() > 1.0:
|
73 |
-
image = image / 255.0
|
74 |
-
|
75 |
-
# Add batch dimension if not present
|
76 |
-
if len(image.shape) == 3:
|
77 |
-
image = np.expand_dims(image, axis=0)
|
78 |
-
|
79 |
-
logger.info(f"Preprocessed image shape: {image.shape}")
|
80 |
-
return image
|
81 |
-
|
82 |
-
except Exception as e:
|
83 |
-
logger.error(f"Error preprocessing image: {str(e)}")
|
84 |
-
raise
|
85 |
|
86 |
-
|
87 |
-
|
88 |
-
|
89 |
-
|
90 |
-
|
91 |
-
|
92 |
-
|
93 |
-
}
|
94 |
-
except Exception as e:
|
95 |
-
logger.error(f"Error during prediction: {str(e)}")
|
96 |
-
raise
|
97 |
-
|
98 |
-
class MedicalDiagnosisAPI:
|
99 |
-
def __init__(self, api_key: str, user_id: str):
|
100 |
-
self.api_key = api_key
|
101 |
-
self.user_id = user_id
|
102 |
-
self.base_url = "https://api.example.com/v1" # Replace with actual API URL
|
103 |
|
104 |
-
|
105 |
-
"""Create a new chat session and return session ID."""
|
106 |
try:
|
107 |
-
|
108 |
-
|
109 |
-
|
110 |
-
"Authorization": f"Bearer {self.api_key}",
|
111 |
-
"X-User-ID": self.user_id
|
112 |
-
}
|
113 |
-
)
|
114 |
-
response.raise_for_status()
|
115 |
-
return response.json()["session_id"]
|
116 |
except Exception as e:
|
117 |
-
logger.error(f"
|
118 |
-
|
119 |
-
|
120 |
-
|
121 |
-
|
122 |
-
|
123 |
-
|
124 |
-
payload = {
|
125 |
-
"patient_info": patient_info,
|
126 |
-
"image_analysis": image_analysis
|
127 |
-
}
|
128 |
|
129 |
-
|
130 |
-
|
131 |
-
|
132 |
-
|
133 |
-
|
134 |
-
|
135 |
-
|
136 |
-
|
137 |
-
|
138 |
-
return response.json()
|
139 |
-
except Exception as e:
|
140 |
-
logger.error(f"Error submitting query: {str(e)}")
|
141 |
-
raise
|
142 |
|
143 |
-
|
144 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
145 |
try:
|
146 |
-
#
|
147 |
-
|
148 |
-
|
149 |
-
|
150 |
-
|
151 |
-
|
152 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
153 |
except Exception as e:
|
154 |
-
logger.error(f"Error
|
155 |
raise
|
156 |
|
157 |
-
|
158 |
-
|
159 |
-
|
160 |
-
|
161 |
-
|
162 |
-
def process_request(self, patient_info: str,
|
163 |
-
image: Optional[np.ndarray]) -> str:
|
164 |
-
"""Process a medical diagnosis request."""
|
165 |
-
try:
|
166 |
-
if self.model.model is None:
|
167 |
-
return json.dumps({
|
168 |
-
"error": "Model initialization failed",
|
169 |
-
"status": "error"
|
170 |
-
}, indent=2)
|
171 |
-
|
172 |
-
# Process image if provided
|
173 |
-
image_analysis = None
|
174 |
-
if image is not None:
|
175 |
-
processed_image = self.model.preprocess_image(image)
|
176 |
-
image_analysis = self.model.predict(processed_image)
|
177 |
-
logger.info(f"Image analysis results: {image_analysis}")
|
178 |
-
|
179 |
-
# Create chat session and submit query
|
180 |
-
session_id = self.api.create_chat_session()
|
181 |
-
llm_response = self.api.submit_query(
|
182 |
-
session_id,
|
183 |
-
patient_info,
|
184 |
-
json.dumps(image_analysis) if image_analysis else None
|
185 |
-
)
|
186 |
-
|
187 |
-
if not llm_response or 'data' not in llm_response or 'answer' not in llm_response['data']:
|
188 |
-
raise ValueError("Invalid response structure from LLM")
|
189 |
-
|
190 |
-
json_data = extract_json_from_answer(llm_response['data']['answer'])
|
191 |
-
return json.dumps(json_data, indent=2)
|
192 |
-
|
193 |
-
except Exception as e:
|
194 |
-
logger.error(f"Error processing request: {str(e)}")
|
195 |
return json.dumps({
|
196 |
-
"error":
|
197 |
-
"status": "error"
|
198 |
-
"details": "Check the application logs for more information"
|
199 |
}, indent=2)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
200 |
|
201 |
-
|
202 |
-
|
203 |
-
|
204 |
-
|
205 |
-
|
206 |
-
|
207 |
-
|
208 |
-
|
209 |
-
|
210 |
-
fn=app.process_request,
|
211 |
-
inputs=[
|
212 |
-
gr.Textbox(
|
213 |
-
label="Patient Information",
|
214 |
-
placeholder="Enter patient details including: symptoms, medical history, current medications, age, gender, and any relevant test results...",
|
215 |
-
lines=5,
|
216 |
-
max_lines=10
|
217 |
-
),
|
218 |
-
gr.Image(
|
219 |
-
label="Medical Image",
|
220 |
-
type="numpy",
|
221 |
-
interactive=True
|
222 |
-
)
|
223 |
-
],
|
224 |
-
outputs=gr.Textbox(
|
225 |
-
label="Medical Analysis",
|
226 |
-
placeholder="JSON analysis will appear here...",
|
227 |
-
lines=15
|
228 |
),
|
229 |
-
|
230 |
-
|
231 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
232 |
|
233 |
if __name__ == "__main__":
|
234 |
-
#
|
235 |
-
logger.info(f"
|
236 |
logger.info(f"TensorFlow Hub version: {hub.__version__}")
|
237 |
logger.info(f"Gradio version: {gr.__version__}")
|
238 |
|
239 |
-
# Create and launch the interface
|
240 |
-
iface = create_gradio_interface()
|
241 |
iface.launch(
|
242 |
server_name="0.0.0.0",
|
243 |
-
share=True, # Enable public link
|
244 |
debug=True
|
245 |
)
|
246 |
|
|
|
7 |
import numpy as np
|
8 |
from PIL import Image
|
9 |
import io
|
10 |
+
import os
|
11 |
|
12 |
+
# Set up logging with more detailed format
|
13 |
logging.basicConfig(
|
14 |
level=logging.INFO,
|
15 |
format='%(asctime)s - %(name)s - %(levelname)s - %(message)s'
|
16 |
)
|
17 |
logger = logging.getLogger(__name__)
|
18 |
|
19 |
+
# API key and user ID for on-demand
|
20 |
+
api_key = 'KGSjxB1uptfSk8I8A7ciCuNT9Xa3qWC3'
|
21 |
+
external_user_id = 'plugin-1717464304'
|
22 |
+
|
23 |
+
def load_model():
|
24 |
+
try:
|
25 |
+
model_path = 'model_epoch_01.h5.keras'
|
26 |
|
27 |
+
# Check if model file exists
|
28 |
+
if not os.path.exists(model_path):
|
29 |
+
raise FileNotFoundError(f"Model file not found at {model_path}")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
30 |
|
31 |
+
logger.info(f"Attempting to load model from {model_path}")
|
32 |
+
|
33 |
+
# Define custom objects dictionary
|
34 |
+
custom_objects = {
|
35 |
+
'KerasLayer': hub.KerasLayer
|
36 |
+
# Add more custom objects if needed
|
37 |
+
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
38 |
|
39 |
+
# Try loading with different configurations
|
|
|
40 |
try:
|
41 |
+
logger.info("Attempting to load model with custom objects...")
|
42 |
+
with tf.keras.utils.custom_object_scope(custom_objects):
|
43 |
+
model = tf_keras.models.load_model(model_path, compile=False)
|
|
|
|
|
|
|
|
|
|
|
|
|
44 |
except Exception as e:
|
45 |
+
logger.error(f"Failed to load with custom objects: {str(e)}")
|
46 |
+
logger.info("Attempting to load model without custom objects...")
|
47 |
+
model = tf_keras.models.load_model(model_path, compile=False)
|
48 |
+
|
49 |
+
# Verify model loaded correctly
|
50 |
+
if model is None:
|
51 |
+
raise ValueError("Model loading returned None")
|
|
|
|
|
|
|
|
|
52 |
|
53 |
+
# Print model summary for debugging
|
54 |
+
model.summary()
|
55 |
+
logger.info("Model loaded successfully")
|
56 |
+
return model
|
57 |
+
|
58 |
+
except Exception as e:
|
59 |
+
logger.error(f"Error loading model: {str(e)}")
|
60 |
+
logger.error(f"Model loading failed with exception type: {type(e)}")
|
61 |
+
raise
|
|
|
|
|
|
|
|
|
62 |
|
63 |
+
# Initialize the model globally
|
64 |
+
try:
|
65 |
+
logger.info("Initializing model...")
|
66 |
+
model = load_model()
|
67 |
+
logger.info("Model initialization completed")
|
68 |
+
except Exception as e:
|
69 |
+
logger.error(f"Failed to initialize model: {str(e)}")
|
70 |
+
model = None
|
71 |
+
|
72 |
+
def preprocess_image(image):
|
73 |
try:
|
74 |
+
# Log image shape and type for debugging
|
75 |
+
logger.info(f"Input image shape: {image.shape}, dtype: {image.dtype}")
|
76 |
+
|
77 |
+
image = image.convert('rgb')
|
78 |
+
image = image.resize((256, 256, 3))
|
79 |
+
image = np.array(image)
|
80 |
+
|
81 |
+
# Normalize pixel values
|
82 |
+
image = image / 255.0
|
83 |
+
|
84 |
+
# Add batch dimension
|
85 |
+
image = np.expand_dims(image, axis=0)
|
86 |
+
logger.info(f"Final preprocessed image shape: {image.shape}")
|
87 |
+
|
88 |
+
return image
|
89 |
+
|
90 |
except Exception as e:
|
91 |
+
logger.error(f"Error preprocessing image: {str(e)}")
|
92 |
raise
|
93 |
|
94 |
+
def gradio_interface(patient_info, image):
|
95 |
+
try:
|
96 |
+
if model is None:
|
97 |
+
logger.error("Model is not initialized")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
98 |
return json.dumps({
|
99 |
+
"error": "Model initialization failed. Please check the logs for details.",
|
100 |
+
"status": "error"
|
|
|
101 |
}, indent=2)
|
102 |
+
|
103 |
+
# Process image if provided
|
104 |
+
image_analysis = None
|
105 |
+
if image is not None:
|
106 |
+
logger.info("Processing uploaded image")
|
107 |
+
# Preprocess image
|
108 |
+
processed_image = preprocess_image(image)
|
109 |
+
|
110 |
+
# Get model prediction
|
111 |
+
logger.info("Running model prediction")
|
112 |
+
prediction = model.predict(processed_image)
|
113 |
+
logger.info(f"Raw prediction shape: {prediction.shape}")
|
114 |
+
logger.info(f"Prediction: {prediction}")
|
115 |
+
# Format prediction results
|
116 |
+
image_analysis = {
|
117 |
+
"prediction": float(prediction[0][0]),
|
118 |
+
"confidence": float(prediction[0][0]) * 100
|
119 |
+
}
|
120 |
+
logger.info(f"Image analysis results: {image_analysis}")
|
121 |
+
|
122 |
+
# Create chat session and submit query
|
123 |
+
session_id = create_chat_session()
|
124 |
+
llm_response = submit_query(session_id, patient_info,
|
125 |
+
json.dumps(image_analysis) if image_analysis else None)
|
126 |
+
|
127 |
+
if not llm_response or 'data' not in llm_response or 'answer' not in llm_response['data']:
|
128 |
+
raise ValueError("Invalid response structure from LLM")
|
129 |
+
|
130 |
+
# Extract and clean JSON from the response
|
131 |
+
json_data = extract_json_from_answer(llm_response['data']['answer'])
|
132 |
+
|
133 |
+
return json.dumps(json_data, indent=2)
|
134 |
+
|
135 |
+
except Exception as e:
|
136 |
+
logger.error(f"Error in gradio_interface: {str(e)}")
|
137 |
+
return json.dumps({
|
138 |
+
"error": str(e),
|
139 |
+
"status": "error",
|
140 |
+
"details": "Check the application logs for more information"
|
141 |
+
}, indent=2)
|
142 |
|
143 |
+
# Gradio interface
|
144 |
+
iface = gr.Interface(
|
145 |
+
fn=gradio_interface,
|
146 |
+
inputs=[
|
147 |
+
gr.Textbox(
|
148 |
+
label="Patient Information",
|
149 |
+
placeholder="Enter patient details including: symptoms, medical history, current medications, age, gender, and any relevant test results...",
|
150 |
+
lines=5,
|
151 |
+
max_lines=10
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
152 |
),
|
153 |
+
gr.Image(
|
154 |
+
label="Medical Image",
|
155 |
+
type="numpy",
|
156 |
+
interactive=True
|
157 |
+
)
|
158 |
+
],
|
159 |
+
outputs=gr.Textbox(
|
160 |
+
label="Medical Analysis",
|
161 |
+
placeholder="JSON analysis will appear here...",
|
162 |
+
lines=15
|
163 |
+
),
|
164 |
+
title="Medical Diagnosis Assistant",
|
165 |
+
description="Enter patient information and optionally upload a medical image for analysis."
|
166 |
+
)
|
167 |
|
168 |
if __name__ == "__main__":
|
169 |
+
# Add version information logging
|
170 |
+
logger.info(f"TensorFlow Keras version: {tf_keras.__version__}")
|
171 |
logger.info(f"TensorFlow Hub version: {hub.__version__}")
|
172 |
logger.info(f"Gradio version: {gr.__version__}")
|
173 |
|
|
|
|
|
174 |
iface.launch(
|
175 |
server_name="0.0.0.0",
|
|
|
176 |
debug=True
|
177 |
)
|
178 |
|