Update app.py

app.py

@@ -1,36 +1,46 @@
 import gradio as gr
-import tensorflow as tf
-import numpy as np
-from PIL import Image as PILImage
 import base64
 from io import BytesIO
+from PIL import Image
+import tensorflow as tf
+import numpy as np
 
+# Carregar o modelo TensorFlow Lite
 interpreter = tf.lite.Interpreter(model_path="model_unquant.tflite")
 interpreter.allocate_tensors()
+
 input_details = interpreter.get_input_details()
 output_details = interpreter.get_output_details()
 classes = ['Bastonete', 'Basófilo']
 
-def recognize_image(base64_image):
-    image_data = base64.b64decode(base64_image)
-    image = PILImage.open(BytesIO(image_data)).convert("RGB")
-    # REMOVER O REDIMENSIONAMENTO
-    # image = image.resize((224, 224))  # Não é mais necessário redimensionar aqui
-    image_array = np.array(image).astype(np.float32)
-    image_array /= 255.0
-    image_input = np.expand_dims(image_array, axis=0)
-    interpreter.set_tensor(input_details[0]['index'], image_input)
-    interpreter.invoke()
-    output_data = interpreter.get_tensor(output_details[0]['index'])
-    predicted_class_index = np.argmax(output_data)
-    predicted_class_name = classes[predicted_class_index]
-    predicted_confidence = output_data[0][predicted_class_index] * 100
+def predict_base64_image(base64_image):
+    try:
+        # Decodificar Base64 para imagem
+        image_data = base64.b64decode(base64_image)
+        image = Image.open(BytesIO(image_data)).convert("RGB")
+        image_array = np.array(image).astype(np.float32) / 255.0
+        image_array = np.expand_dims(image_array, axis=0)
 
-    return f"Classe: {predicted_class_name} (Confiança: {predicted_confidence:.2f}%)"
+        # Realizar a inferência
+        interpreter.set_tensor(input_details[0]['index'], image_array)
+        interpreter.invoke()
+        output_data = interpreter.get_tensor(output_details[0]['index'])
+        predicted_class_index = np.argmax(output_data)
+        predicted_class_name = classes[predicted_class_index]
+        predicted_confidence = output_data[0][predicted_class_index] * 100
 
+        # Retornar o resultado
+        return {"class": predicted_class_name, "confidence": f"{predicted_confidence:.2f}%"}
+    except Exception as e:
+        return {"error": str(e)}
 
+# Configuração do Gradio
 interface = gr.Interface(
-    fn=recognize_image,
-    inputs="text",
-    outputs="text"
-)
+    fn=predict_base64_image,
+    inputs="text",  # Base64 será enviado como texto
+    outputs="json",  # Retorna um JSON com a classe e a confiança
+    api_name="/predict"
+)
+
+# Iniciar o servidor
+interface.launch()