Commit
路
4f8e957
1
Parent(s):
6317a79
Edit app.py
Browse files
app.py
CHANGED
|
@@ -35,15 +35,11 @@ def make_gradcam_heatmap(img_array, model, last_conv_layer_name="conv4", pred_in
|
|
| 35 |
)
|
| 36 |
with tf.GradientTape() as tape:
|
| 37 |
conv_outputs, predictions = grad_model(img_array)
|
| 38 |
-
|
| 39 |
-
# 馃憞 Aseguramos que predictions sea tensor
|
| 40 |
if isinstance(predictions, list):
|
| 41 |
predictions = predictions[0]
|
| 42 |
-
|
| 43 |
if pred_index is None:
|
| 44 |
pred_index = tf.argmax(predictions[0])
|
| 45 |
class_channel = predictions[:, pred_index]
|
| 46 |
-
|
| 47 |
grads = tape.gradient(class_channel, conv_outputs)
|
| 48 |
pooled_grads = tf.reduce_mean(grads, axis=(0, 1, 2))
|
| 49 |
conv_outputs = conv_outputs[0]
|
|
@@ -52,48 +48,40 @@ def make_gradcam_heatmap(img_array, model, last_conv_layer_name="conv4", pred_in
|
|
| 52 |
heatmap = tf.maximum(heatmap, 0) / tf.math.reduce_max(heatmap)
|
| 53 |
return heatmap.numpy()
|
| 54 |
|
| 55 |
-
# === Funci贸n
|
| 56 |
def preprocess_and_predict(img_input):
|
| 57 |
-
# Convertir PIL a OpenCV (BGR)
|
| 58 |
img = np.array(img_input)[:, :, ::-1]
|
| 59 |
-
|
| 60 |
-
# 1. Zoom
|
| 61 |
zoomed = apply_zoom(img, zoom_factor=0.9)
|
| 62 |
-
|
| 63 |
-
# 2. Quitar pelos
|
| 64 |
rgb_clean = cv2.cvtColor(zoomed, cv2.COLOR_BGR2RGB)
|
| 65 |
clean = quitar_pelos(rgb_clean)
|
| 66 |
|
| 67 |
-
# 3. Segmentar
|
| 68 |
mask, lesion_rgb = segmentar_lesion(clean, size=(ROWS, COLS))
|
| 69 |
|
| 70 |
-
# 4. Preparar para SimpleNet
|
| 71 |
lesion_resized = cv2.resize(lesion_rgb, (ROWS, COLS))
|
| 72 |
img_array = image.img_to_array(lesion_resized) / 255.0
|
| 73 |
img_array = np.expand_dims(img_array, axis=0)
|
| 74 |
|
| 75 |
-
# 5. Predicci贸n con SimpleNet
|
| 76 |
probs = model.predict(img_array)[0]
|
| 77 |
classes = ["Benign", "Malignant"]
|
| 78 |
pred_idx = np.argmax(probs)
|
| 79 |
pred_label = classes[pred_idx]
|
| 80 |
result_text = f"Predicci贸n: {pred_label} ({probs[pred_idx]*100:.2f}%)"
|
| 81 |
|
| 82 |
-
#
|
| 83 |
area = calcular_area(mask)
|
| 84 |
perim = calcular_perimetro(mask)
|
| 85 |
circ = calcular_circularidad(mask)
|
| 86 |
sim_v, sim_h = calcular_simetria(mask)
|
| 87 |
|
| 88 |
-
|
| 89 |
-
|
| 90 |
-
|
| 91 |
-
|
| 92 |
-
|
| 93 |
-
|
| 94 |
-
|
| 95 |
|
| 96 |
-
#
|
| 97 |
raw_resized = cv2.resize(np.array(img_input), (ROWS, COLS))
|
| 98 |
raw_array = image.img_to_array(raw_resized) / 255.0
|
| 99 |
raw_array = np.expand_dims(raw_array, axis=0)
|
|
@@ -104,22 +92,110 @@ def preprocess_and_predict(img_input):
|
|
| 104 |
heatmap_color = cv2.applyColorMap(heatmap, cv2.COLORMAP_JET)
|
| 105 |
overlay = cv2.addWeighted(raw_resized.astype("uint8"), 0.6, heatmap_color, 0.4, 0)
|
| 106 |
|
| 107 |
-
return mask, lesion_rgb, result_text,
|
| 108 |
-
|
| 109 |
-
|
| 110 |
-
|
| 111 |
-
|
| 112 |
-
|
| 113 |
-
|
| 114 |
-
|
| 115 |
-
|
| 116 |
-
|
| 117 |
-
|
| 118 |
-
|
| 119 |
-
|
| 120 |
-
|
| 121 |
-
|
| 122 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 123 |
|
|
|
|
| 124 |
if __name__ == "__main__":
|
| 125 |
demo.launch()
|
|
|
|
| 35 |
)
|
| 36 |
with tf.GradientTape() as tape:
|
| 37 |
conv_outputs, predictions = grad_model(img_array)
|
|
|
|
|
|
|
| 38 |
if isinstance(predictions, list):
|
| 39 |
predictions = predictions[0]
|
|
|
|
| 40 |
if pred_index is None:
|
| 41 |
pred_index = tf.argmax(predictions[0])
|
| 42 |
class_channel = predictions[:, pred_index]
|
|
|
|
| 43 |
grads = tape.gradient(class_channel, conv_outputs)
|
| 44 |
pooled_grads = tf.reduce_mean(grads, axis=(0, 1, 2))
|
| 45 |
conv_outputs = conv_outputs[0]
|
|
|
|
| 48 |
heatmap = tf.maximum(heatmap, 0) / tf.math.reduce_max(heatmap)
|
| 49 |
return heatmap.numpy()
|
| 50 |
|
| 51 |
+
# === Funci贸n principal ===
|
| 52 |
def preprocess_and_predict(img_input):
|
|
|
|
| 53 |
img = np.array(img_input)[:, :, ::-1]
|
|
|
|
|
|
|
| 54 |
zoomed = apply_zoom(img, zoom_factor=0.9)
|
|
|
|
|
|
|
| 55 |
rgb_clean = cv2.cvtColor(zoomed, cv2.COLOR_BGR2RGB)
|
| 56 |
clean = quitar_pelos(rgb_clean)
|
| 57 |
|
|
|
|
| 58 |
mask, lesion_rgb = segmentar_lesion(clean, size=(ROWS, COLS))
|
| 59 |
|
|
|
|
| 60 |
lesion_resized = cv2.resize(lesion_rgb, (ROWS, COLS))
|
| 61 |
img_array = image.img_to_array(lesion_resized) / 255.0
|
| 62 |
img_array = np.expand_dims(img_array, axis=0)
|
| 63 |
|
|
|
|
| 64 |
probs = model.predict(img_array)[0]
|
| 65 |
classes = ["Benign", "Malignant"]
|
| 66 |
pred_idx = np.argmax(probs)
|
| 67 |
pred_label = classes[pred_idx]
|
| 68 |
result_text = f"Predicci贸n: {pred_label} ({probs[pred_idx]*100:.2f}%)"
|
| 69 |
|
| 70 |
+
# === M茅tricas geom茅tricas en formato tabla ===
|
| 71 |
area = calcular_area(mask)
|
| 72 |
perim = calcular_perimetro(mask)
|
| 73 |
circ = calcular_circularidad(mask)
|
| 74 |
sim_v, sim_h = calcular_simetria(mask)
|
| 75 |
|
| 76 |
+
metrics_data = [
|
| 77 |
+
["脕rea (px虏)", area],
|
| 78 |
+
["Per铆metro (px)", perim],
|
| 79 |
+
["Circularidad", round(circ, 3)],
|
| 80 |
+
["Simetr铆a Vertical", round(sim_v, 3)],
|
| 81 |
+
["Simetr铆a Horizontal", round(sim_h, 3)]
|
| 82 |
+
]
|
| 83 |
|
| 84 |
+
# === Grad-CAM ===
|
| 85 |
raw_resized = cv2.resize(np.array(img_input), (ROWS, COLS))
|
| 86 |
raw_array = image.img_to_array(raw_resized) / 255.0
|
| 87 |
raw_array = np.expand_dims(raw_array, axis=0)
|
|
|
|
| 92 |
heatmap_color = cv2.applyColorMap(heatmap, cv2.COLORMAP_JET)
|
| 93 |
overlay = cv2.addWeighted(raw_resized.astype("uint8"), 0.6, heatmap_color, 0.4, 0)
|
| 94 |
|
| 95 |
+
return mask, lesion_rgb, result_text, metrics_data, overlay
|
| 96 |
+
|
| 97 |
+
|
| 98 |
+
# === Interfaz con estilo ===
|
| 99 |
+
with gr.Blocks(css="""
|
| 100 |
+
body, .gradio-container {
|
| 101 |
+
font-family: 'Inter', sans-serif;
|
| 102 |
+
background: #ffffff !important;
|
| 103 |
+
}
|
| 104 |
+
h1, h2 { font-weight: 600; color: #111827; margin-bottom: 0.5rem; }
|
| 105 |
+
.section { background: white; border-radius: 0.75rem; padding: 1.5rem; margin: 1rem auto;
|
| 106 |
+
box-shadow: 0 1px 4px rgba(0,0,0,0.05); max-width: 800px; }
|
| 107 |
+
.gradio-container { max-width: 900px; margin: auto; }
|
| 108 |
+
img { border-radius: 0.5rem; }
|
| 109 |
+
#analyze-btn {
|
| 110 |
+
display: block;
|
| 111 |
+
margin: 1rem auto;
|
| 112 |
+
width: 150px;
|
| 113 |
+
background-color: #374151;
|
| 114 |
+
color: white;
|
| 115 |
+
font-weight: bold;
|
| 116 |
+
}
|
| 117 |
+
#metrics-table {
|
| 118 |
+
max-width: 320px;
|
| 119 |
+
margin: 1rem auto;
|
| 120 |
+
}
|
| 121 |
+
.expl-text {
|
| 122 |
+
color: #4B5563;
|
| 123 |
+
font-size: 0.95rem;
|
| 124 |
+
margin-bottom: 1rem;
|
| 125 |
+
}
|
| 126 |
+
""") as demo:
|
| 127 |
+
|
| 128 |
+
# === T铆tulo e introducci贸n ===
|
| 129 |
+
gr.HTML("""
|
| 130 |
+
<section style="text-align:center; padding: 2rem;">
|
| 131 |
+
<h1>DermaScan - Clasificaci贸n de Lesiones</h1>
|
| 132 |
+
<p style="color:#374151; font-size:1.05rem; max-width: 700px; margin:auto;">
|
| 133 |
+
El melanoma es uno de los c谩nceres de piel m谩s agresivos.
|
| 134 |
+
Este sistema utiliza <b>Inteligencia Artificial</b> para apoyar el diagn贸stico,
|
| 135 |
+
mostrando resultados de clasificaci贸n, zonas relevantes (Grad-CAM)
|
| 136 |
+
y m茅tricas geom茅tricas basadas en el criterio ABCDE.
|
| 137 |
+
</p>
|
| 138 |
+
</section>
|
| 139 |
+
""")
|
| 140 |
+
|
| 141 |
+
# === Subir imagen ===
|
| 142 |
+
with gr.Column(elem_classes="section"):
|
| 143 |
+
gr.HTML("<h2>Subir imagen</h2>")
|
| 144 |
+
gr.HTML('<p class="expl-text">Sube una imagen dermatosc贸pica de la lesi贸n para analizarla.</p>')
|
| 145 |
+
img_input = gr.Image(type="pil", label="Imagen de la lesi贸n", elem_id="upload-img")
|
| 146 |
+
run_btn = gr.Button("Analizar", elem_id="analyze-btn")
|
| 147 |
+
|
| 148 |
+
# === Segmentaci贸n ===
|
| 149 |
+
with gr.Column(elem_classes="section"):
|
| 150 |
+
gr.HTML("<h2>Preprocesamiento y Segmentaci贸n</h2>")
|
| 151 |
+
gr.HTML("""
|
| 152 |
+
<p class="expl-text">
|
| 153 |
+
En este paso se realizan varias operaciones:
|
| 154 |
+
<br>- Conversi贸n de canales de color.<br>
|
| 155 |
+
- Eliminaci贸n de pelos.<br>
|
| 156 |
+
- Segmentaci贸n de la lesi贸n.<br>
|
| 157 |
+
</p>
|
| 158 |
+
""")
|
| 159 |
+
img_mask = gr.Image(type="numpy", label="M谩scara Binaria", elem_id="mask-img")
|
| 160 |
+
img_segmented = gr.Image(type="numpy", label="Lesi贸n Segmentada", elem_id="seg-img")
|
| 161 |
+
|
| 162 |
+
# === Grad-CAM ===
|
| 163 |
+
with gr.Column(elem_classes="section"):
|
| 164 |
+
gr.HTML("<h2>Grad-CAM (ZoomNet)</h2>")
|
| 165 |
+
gr.HTML("""
|
| 166 |
+
<p class="expl-text">
|
| 167 |
+
El mapa de calor muestra las zonas con mayor relevancia para el modelo
|
| 168 |
+
al determinar si la lesi贸n es benigna o maligna.
|
| 169 |
+
</p>
|
| 170 |
+
""")
|
| 171 |
+
gradcam_img = gr.Image(type="numpy", label="Mapa de activaci贸n", elem_id="gradcam-img")
|
| 172 |
+
|
| 173 |
+
# === Resultados ===
|
| 174 |
+
with gr.Column(elem_classes="section"):
|
| 175 |
+
gr.HTML("<h2>Resultados del modelo</h2>")
|
| 176 |
+
result_text = gr.Textbox(label="Clasificaci贸n", max_lines=1, interactive=False)
|
| 177 |
+
|
| 178 |
+
gr.HTML("""
|
| 179 |
+
<p class="expl-text">
|
| 180 |
+
M茅tricas geom茅tricas basadas en el criterio <b>ABCDE</b>:
|
| 181 |
+
煤tiles para analizar lesiones en casos dudosos.
|
| 182 |
+
</p>
|
| 183 |
+
""")
|
| 184 |
+
metrics_table = gr.Dataframe(
|
| 185 |
+
headers=["M茅trica", "Valor"],
|
| 186 |
+
datatype=["str", "number"],
|
| 187 |
+
interactive=False,
|
| 188 |
+
label="M茅tricas calculadas",
|
| 189 |
+
elem_id="metrics-table"
|
| 190 |
+
)
|
| 191 |
+
|
| 192 |
+
# === Conexi贸n bot贸n -> funci贸n ===
|
| 193 |
+
run_btn.click(
|
| 194 |
+
fn=preprocess_and_predict,
|
| 195 |
+
inputs=[img_input],
|
| 196 |
+
outputs=[img_mask, img_segmented, result_text, metrics_table, gradcam_img]
|
| 197 |
+
)
|
| 198 |
|
| 199 |
+
# === Lanzar en tema claro ===
|
| 200 |
if __name__ == "__main__":
|
| 201 |
demo.launch()
|