Update app.py

app.py

@@ -2,48 +2,43 @@ import gradio as gr
 import numpy as np
 from tensorflow.keras.preprocessing.image import img_to_array
 from tensorflow.keras.models import load_model
-from PIL import Image
+from PIL import Image, ImageOps
+import json
 
-# --- Constants ---
-IMG_SIZE = (224, 224)
-# antes: CLASS_ORDER = [ … ]
-CLASS_ORDER = sorted(['Healthy', 'Leaf Blight', 'Black Rot', 'ESCA'])
-MODEL_PATH = "cropvision_model.keras"  # ensure this file is in the same folder
+# Constantes (mantém IMG_SIZE, CLASS_ORDER, etc)
+IMG_SIZE    = (224, 224)
+CLASS_ORDER = sorted(['Healthy','Leaf Blight','Black Rot','ESCA'])  
+# → ['Black Rot','ESCA','Healthy','Leaf Blight']
 
+model = load_model("cropvision_model.keras")
 
-# --- Load model once at startup ---
-model = load_model(MODEL_PATH)
-import hashlib, pathlib, json
-
-# --- debug ---
-h = hashlib.md5(pathlib.Path(MODEL_PATH).read_bytes()).hexdigest()
-print("📝 MD5 do modelo:", h)
-print("Camadas:", len(model.layers), "| Params:", model.count_params())
-print("class_indices que o generator gravou:", json.dumps(model.class_indices if hasattr(model, 'class_indices') else "N/A"))
-# -------------
-
-
-# --- Prediction function ---
 def predict(image: Image.Image):
-    # Resize & preprocess
-    img = image.convert("RGB").resize(IMG_SIZE)
-    arr = img_to_array(img) / 255.0
-    arr = np.expand_dims(arr, 0)
+    # pré-processamento robusto
+    img = ImageOps.exif_transpose(image).convert("RGB")
+    img = ImageOps.fit(img, IMG_SIZE, Image.Resampling.LANCZOS)
+
+    # inferência
+    arr   = img_to_array(img) / 255.0
+    arr   = np.expand_dims(arr, 0)
+    probs = model.predict(arr)[0]         # vetor de 4 probabilidades
 
-    # Inference
-    probs = model.predict(arr)
-    cls   = CLASS_ORDER[int(np.argmax(probs))]
-    conf  = float(np.max(probs))
+    idx   = int(np.argmax(probs))
+    cls   = CLASS_ORDER[idx]
+    # mapeamento label → probabilidade
+    mapping = {label: float(probs[i]) for i,label in enumerate(CLASS_ORDER)}
 
-    return f"{cls}  ({conf:.1%})"
+    # devolve (rótulo, JSON-formatado das probabilidades)
+    return cls, json.dumps(mapping, indent=2)
 
-# --- Gradio UI ---
 demo = gr.Interface(
     fn=predict,
-    inputs=gr.Image(type="pil", label="Carrega uma imagem"),
-    outputs=gr.Textbox(label="Resultado"),
-    title="CropVision – classificação de doenças nas vinhas",
-    description="Modelo que vai distinguir Healthy / Leaf Blight / Black Rot / ESCA"
+    inputs=gr.Image(type="pil", label="Carrega uma folha"),
+    outputs=[
+      gr.Textbox(label="Classe predita"),
+      gr.Code(label="Probabilidades (label:valor)")
+    ],
+    title="CropVision – classificação de folhas",
+    description="Healthy / Leaf Blight / Black Rot / ESCA"
 )
 
 if __name__ == "__main__":