dcavadia commited on
Commit
62ba5b3
·
verified ·
1 Parent(s): 6acfdde

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +97 -159
app.py CHANGED
@@ -4,14 +4,13 @@ import gradio as gr
4
  import onnxruntime as ort
5
  from PIL import Image
6
  from torchvision import transforms
7
- import pandas as pd
8
  import time
9
  import os
10
 
11
  # ----------------------------
12
  # Model + metadata
13
  # ----------------------------
14
- ORT_PROVIDERS = ["CPUExecutionProvider"] # add "CUDAExecutionProvider" if available
15
 
16
  try:
17
  ort_session = ort.InferenceSession("model_new_new_final.onnx", providers=ORT_PROVIDERS)
@@ -23,7 +22,7 @@ except Exception as e:
23
  try:
24
  with open("dat.json", "r", encoding="utf-8") as f:
25
  data = json.load(f)
26
- CLASSES = list(data) # ordered list of class names
27
  print(f"Loaded {len(CLASSES)} classes from JSON")
28
  except Exception as e:
29
  print(f"Error loading JSON: {e}")
@@ -49,38 +48,15 @@ def probabilities_to_ints(probabilities, total_sum=100):
49
  rounded = rounded.reshape(scaled.shape)
50
  return rounded
51
 
52
- def create_prob_table(classes, probabilities):
53
- """Create a formatted HTML table for probabilities"""
54
- df = pd.DataFrame({
55
- "Enfermedad": classes,
56
- "Probabilidad (%)": probabilities
57
- }).sort_values("Probabilidad (%)", ascending=False)
58
 
59
- # Create HTML table
60
- html = """
61
- <div style='max-height: 300px; overflow-y: auto;'>
62
- <table style='width: 100%; border-collapse: collapse; font-family: Arial, sans-serif;'>
63
- <thead style='background-color: #f8f9fa; position: sticky; top: 0;'>
64
- <tr>
65
- <th style='padding: 8px; text-align: left; border: 1px solid #dee2e6;'>Enfermedad</th>
66
- <th style='padding: 8px; text-align: right; border: 1px solid #dee2e6;'>Probabilidad (%)</th>
67
- </tr>
68
- </thead>
69
- <tbody>
70
- """
71
 
72
- for _, row in df.iterrows():
73
- prob = int(row["Probabilidad (%)"])
74
- color = "#dc3545" if prob > 50 else "#28a745" if prob > 20 else "#6c757d"
75
- html += f"""
76
- <tr>
77
- <td style='padding: 8px; border: 1px solid #dee2e6;'>{row['Enfermedad']}</td>
78
- <td style='padding: 8px; text-align: right; border: 1px solid #dee2e6; color: {color}; font-weight: bold;'>{prob}%</td>
79
- </tr>
80
- """
81
-
82
- html += "</tbody></table></div>"
83
- return html
84
 
85
  MEAN = [0.7611, 0.5869, 0.5923]
86
  STD = [0.1266, 0.1487, 0.1619]
@@ -97,33 +73,27 @@ def preprocess(pil_img: Image.Image):
97
  # Inference function
98
  # ----------------------------
99
  def predict(image):
100
- print(f"Predict function called with image type: {type(image)}")
101
 
102
- # Check if model and data are loaded
103
  if ort_session is None:
104
- return ("Error: Modelo no cargado", "", "", "", "", "", "Error en el modelo", "Error")
105
 
106
  if not CLASSES:
107
- return ("Error: Datos no cargados", "", "", "", "", "", "Error en los datos", "Error")
108
 
109
- # Handle clicks with no image gracefully
110
  if image is None:
111
- print("No image provided")
112
- return ("Cargue una imagen y presione Analizar.", "", "", "", "", "", "No hay imagen cargada", "")
113
 
114
  try:
115
  if isinstance(image, Image.Image):
116
  pil = image.convert("RGB")
117
  else:
118
- # Handle numpy array from Gradio
119
  pil = Image.fromarray(image.astype('uint8')).convert("RGB")
120
 
121
- print(f"Image processed, size: {pil.size}")
122
 
123
  t0 = time.time()
124
  input_tensor = preprocess(pil)
125
- print(f"Input tensor shape: {input_tensor.shape}")
126
-
127
  input_name = ort_session.get_inputs()[0].name
128
  output = ort_session.run(None, {input_name: input_tensor})
129
 
@@ -131,15 +101,13 @@ def predict(image):
131
  pred_idx = int(np.argmax(logits))
132
  pred_name = CLASSES[pred_idx]
133
 
134
- print(f"Prediction: {pred_name} (index: {pred_idx})")
135
-
136
  # Softmax probabilities
137
  exp = np.exp(logits - np.max(logits))
138
  probs = exp / exp.sum()
139
  conf_text = f"{float(probs[pred_idx]) * 100:.1f}%"
140
 
141
  ints = probabilities_to_ints(probs * 100.0, total_sum=100)
142
- prob_table = create_prob_table(CLASSES, ints)
143
 
144
  details = data[pred_name]
145
  descripcion = details.get("description", "")
@@ -149,130 +117,100 @@ def predict(image):
149
 
150
  latency_ms = int((time.time() - t0) * 1000)
151
 
152
- print(f"Prediction completed in {latency_ms}ms")
153
 
154
- return (pred_name, conf_text, descripcion, sintomas, causas, tratamiento, prob_table, f"{latency_ms} ms")
155
 
156
  except Exception as e:
157
  print(f"Error during prediction: {e}")
158
  import traceback
159
  traceback.print_exc()
160
- return (f"Error: {str(e)}", "", "", "", "", "", "Error en la predicción", "Error")
161
 
162
  # ----------------------------
163
- # Theme (compatible across Gradio versions)
164
  # ----------------------------
165
- try:
166
- theme = gr.themes.Soft(primary_hue="rose", secondary_hue="slate")
167
- except Exception:
168
- theme = None # fallback to default theme
169
-
170
- # CSS polish
171
- CUSTOM_CSS = """
172
- .header {display:flex; align-items:center; gap:12px;}
173
- .badge {font-size:12px; padding:4px 8px; border-radius:12px; background:#f1f5f9; color:#334155;}
174
- .pred-card {font-size:18px;}
175
- .footer {font-size:12px; color:#64748b; text-align:center; padding:12px 0;}
176
- button, .gradio-container .gr-box, .gradio-container .gr-panel { border-radius: 10px !important; }
177
- """
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
178
 
179
  # ----------------------------
180
- # UI
181
  # ----------------------------
182
- with gr.Blocks(theme=theme, css=CUSTOM_CSS) as demo:
183
- with gr.Row():
184
- with gr.Column(scale=6):
185
- gr.Markdown(
186
- """
187
- <div class="header">
188
- <h1 style="margin:0;">Clasificación de Enfermedades de la Piel</h1>
189
- <span class="badge">Demo • No es diagnóstico médico</span>
190
- </div>
191
- <p style="margin-top:6px;">
192
- Sube una imagen dermatoscópica para ver la clase predicha, la confianza y la distribución de probabilidades.
193
- </p>
194
- """
195
- )
196
- with gr.Column(scale=1, min_width=140):
197
- try:
198
- dark_toggle = gr.ThemeMode(label="Modo", value="system")
199
- except Exception:
200
- gr.Markdown("")
201
-
202
- with gr.Row(equal_height=True):
203
- # Left column: input + actions
204
- with gr.Column(scale=5):
205
- image = gr.Image(type="numpy", label="Imagen de la lesión", height=420, sources=["upload", "clipboard"])
206
- with gr.Row():
207
- analyze_btn = gr.Button("Analizar", variant="primary")
208
- clear_btn = gr.Button("Limpiar")
209
-
210
- example_paths = [
211
- "examples/ak.jpg",
212
- "examples/bcc.jpg",
213
- "examples/df.jpg",
214
- "examples/melanoma.jpg",
215
- "examples/nevus.jpg",
216
- ]
217
- example_paths = [p for p in example_paths if os.path.exists(p)]
218
- if example_paths:
219
- gr.Examples(examples=example_paths, inputs=image, label="Ejemplos rápidos")
220
- latency = gr.Label(label="Latencia aproximada")
221
-
222
- # Right column: results
223
- with gr.Column(scale=5):
224
- with gr.Group():
225
- with gr.Row():
226
- nombre = gr.Label(label="Predicción principal", elem_classes=["pred-card"])
227
- confianza = gr.Label(label="Confianza")
228
- # Probability table instead of BarPlot
229
- prob_table = gr.HTML(
230
- value="<p style='text-align: center; color: #6c757d;'>Cargue una imagen y haga clic en 'Analizar' para ver la distribución de probabilidades.</p>",
231
- label="Distribución de Probabilidades"
232
- )
233
- with gr.Tabs():
234
- with gr.TabItem("Detalles"):
235
- with gr.Accordion("Descripción", open=True):
236
- descripcion = gr.Textbox(lines=4, interactive=False)
237
- with gr.Accordion("Síntomas", open=False):
238
- sintomas = gr.Textbox(lines=4, interactive=False)
239
- with gr.Accordion("Causas", open=False):
240
- causas = gr.Textbox(lines=4, interactive=False)
241
- with gr.Accordion("Tratamiento", open=False):
242
- tratamiento = gr.Textbox(lines=4, interactive=False)
243
- with gr.TabItem("Acerca del modelo"):
244
- gr.Markdown(
245
- "- Arquitectura: CNN exportado a ONNX.<br>"
246
- "- Entrenamiento: dataset dermatoscópico (ver documentación).<br>"
247
- "- Nota: Esta herramienta es solo con fines educativos y no reemplaza una evaluación médica."
248
- )
249
-
250
- gr.Markdown("<div class='footer'>Versión del modelo: 1.0 • Última actualización: 2025‑08 • Universidad Central de Venezuela</div>")
251
-
252
- # ----------------------------
253
- # Event handlers
254
- # ----------------------------
255
- outputs = [nombre, confianza, descripcion, sintomas, causas, tratamiento, prob_table, latency]
256
-
257
- # Analyze click runs prediction
258
- analyze_btn.click(
259
- fn=predict,
260
- inputs=[image],
261
- outputs=outputs
262
- )
263
-
264
- # Clear resets input and outputs
265
- def clear_all():
266
- print("Clear button clicked")
267
- empty_table = "<p style='text-align: center; color: #6c757d;'>Cargue una imagen y haga clic en 'Analizar' para ver la distribución de probabilidades.</p>"
268
- return (None, "", "", "", "", "", "", empty_table, "")
269
-
270
- clear_btn.click(
271
- fn=clear_all,
272
- inputs=None,
273
- outputs=[image] + outputs
274
- )
275
-
276
  if __name__ == "__main__":
 
 
277
  print("Starting Gradio demo...")
278
  demo.launch(debug=True)
 
4
  import onnxruntime as ort
5
  from PIL import Image
6
  from torchvision import transforms
 
7
  import time
8
  import os
9
 
10
  # ----------------------------
11
  # Model + metadata
12
  # ----------------------------
13
+ ORT_PROVIDERS = ["CPUExecutionProvider"]
14
 
15
  try:
16
  ort_session = ort.InferenceSession("model_new_new_final.onnx", providers=ORT_PROVIDERS)
 
22
  try:
23
  with open("dat.json", "r", encoding="utf-8") as f:
24
  data = json.load(f)
25
+ CLASSES = list(data)
26
  print(f"Loaded {len(CLASSES)} classes from JSON")
27
  except Exception as e:
28
  print(f"Error loading JSON: {e}")
 
48
  rounded = rounded.reshape(scaled.shape)
49
  return rounded
50
 
51
+ def format_probabilities(classes, probabilities):
52
+ """Create a simple text format for probabilities"""
53
+ result = []
54
+ sorted_items = sorted(zip(classes, probabilities), key=lambda x: x[1], reverse=True)
 
 
55
 
56
+ for disease, prob in sorted_items[:5]: # Show top 5
57
+ result.append(f"{disease}: {int(prob)}%")
 
 
 
 
 
 
 
 
 
 
58
 
59
+ return "\n".join(result)
 
 
 
 
 
 
 
 
 
 
 
60
 
61
  MEAN = [0.7611, 0.5869, 0.5923]
62
  STD = [0.1266, 0.1487, 0.1619]
 
73
  # Inference function
74
  # ----------------------------
75
  def predict(image):
76
+ print(f"Predict function called with image: {image is not None}")
77
 
 
78
  if ort_session is None:
79
+ return "Error: Modelo no cargado", "", "", "", "", "", "Error", ""
80
 
81
  if not CLASSES:
82
+ return "Error: Datos no cargados", "", "", "", "", "", "Error", ""
83
 
 
84
  if image is None:
85
+ return "Cargue una imagen y presione Analizar.", "", "", "", "", "", "No hay imagen", ""
 
86
 
87
  try:
88
  if isinstance(image, Image.Image):
89
  pil = image.convert("RGB")
90
  else:
 
91
  pil = Image.fromarray(image.astype('uint8')).convert("RGB")
92
 
93
+ print(f"Processing image of size: {pil.size}")
94
 
95
  t0 = time.time()
96
  input_tensor = preprocess(pil)
 
 
97
  input_name = ort_session.get_inputs()[0].name
98
  output = ort_session.run(None, {input_name: input_tensor})
99
 
 
101
  pred_idx = int(np.argmax(logits))
102
  pred_name = CLASSES[pred_idx]
103
 
 
 
104
  # Softmax probabilities
105
  exp = np.exp(logits - np.max(logits))
106
  probs = exp / exp.sum()
107
  conf_text = f"{float(probs[pred_idx]) * 100:.1f}%"
108
 
109
  ints = probabilities_to_ints(probs * 100.0, total_sum=100)
110
+ prob_text = format_probabilities(CLASSES, ints)
111
 
112
  details = data[pred_name]
113
  descripcion = details.get("description", "")
 
117
 
118
  latency_ms = int((time.time() - t0) * 1000)
119
 
120
+ print(f"Prediction: {pred_name} with confidence {conf_text}")
121
 
122
+ return pred_name, conf_text, descripcion, sintomas, causas, tratamiento, prob_text, f"{latency_ms} ms"
123
 
124
  except Exception as e:
125
  print(f"Error during prediction: {e}")
126
  import traceback
127
  traceback.print_exc()
128
+ return f"Error: {str(e)}", "", "", "", "", "", "Error en predicción", "Error"
129
 
130
  # ----------------------------
131
+ # Simple UI with minimal components
132
  # ----------------------------
133
+ def create_interface():
134
+ with gr.Blocks(title="Clasificación de Enfermedades de la Piel") as demo:
135
+
136
+ gr.Markdown("""
137
+ # Clasificación de Enfermedades de la Piel
138
+ **Demo • No es diagnóstico médico**
139
+
140
+ Sube una imagen dermatoscópica para ver la clase predicha, la confianza y la distribución de probabilidades.
141
+ """)
142
+
143
+ with gr.Row():
144
+ # Left column
145
+ with gr.Column():
146
+ image_input = gr.Image(type="numpy", label="Imagen de la lesión")
147
+
148
+ with gr.Row():
149
+ analyze_btn = gr.Button("Analizar", variant="primary")
150
+ clear_btn = gr.Button("Limpiar")
151
+
152
+ # Examples if they exist
153
+ example_paths = [
154
+ "examples/ak.jpg", "examples/bcc.jpg", "examples/df.jpg",
155
+ "examples/melanoma.jpg", "examples/nevus.jpg"
156
+ ]
157
+ existing_examples = [p for p in example_paths if os.path.exists(p)]
158
+ if existing_examples:
159
+ gr.Examples(examples=existing_examples, inputs=image_input)
160
+
161
+ # Right column
162
+ with gr.Column():
163
+ prediction_output = gr.Textbox(label="Predicción Principal", interactive=False)
164
+ confidence_output = gr.Textbox(label="Confianza", interactive=False)
165
+ probabilities_output = gr.Textbox(label="Top 5 Probabilidades", interactive=False, lines=5)
166
+ latency_output = gr.Textbox(label="Latencia", interactive=False)
167
+
168
+ # Details section
169
+ with gr.Row():
170
+ with gr.Column():
171
+ gr.Markdown("### Detalles de la Condición")
172
+ description_output = gr.Textbox(label="Descripción", interactive=False, lines=3)
173
+ symptoms_output = gr.Textbox(label="Síntomas", interactive=False, lines=3)
174
+ causes_output = gr.Textbox(label="Causas", interactive=False, lines=3)
175
+ treatment_output = gr.Textbox(label="Tratamiento", interactive=False, lines=3)
176
+
177
+ gr.Markdown("""
178
+ ---
179
+ **Nota importante:** Esta herramienta es solo con fines educativos y de investigación.
180
+ No reemplaza una evaluación médica profesional.
181
+
182
+ *Versión del modelo: 1.0 • Universidad Central de Venezuela*
183
+ """)
184
+
185
+ # Event handlers
186
+ all_outputs = [
187
+ prediction_output, confidence_output, description_output,
188
+ symptoms_output, causes_output, treatment_output,
189
+ probabilities_output, latency_output
190
+ ]
191
+
192
+ analyze_btn.click(
193
+ fn=predict,
194
+ inputs=[image_input],
195
+ outputs=all_outputs
196
+ )
197
+
198
+ def clear_all():
199
+ return None, "", "", "", "", "", "", "", ""
200
+
201
+ clear_btn.click(
202
+ fn=clear_all,
203
+ inputs=None,
204
+ outputs=[image_input] + all_outputs
205
+ )
206
+
207
+ return demo
208
 
209
  # ----------------------------
210
+ # Launch
211
  # ----------------------------
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
212
  if __name__ == "__main__":
213
+ print("Creating interface...")
214
+ demo = create_interface()
215
  print("Starting Gradio demo...")
216
  demo.launch(debug=True)