Update src/app.py

src/app.py

@@ -1,89 +1,68 @@
-import streamlit as st
-import tensorflow as tf
-import numpy as np
-from PIL import Image
-
-# ====== MODEL SETTINGS ======
-MODEL_PATH = "cnn_largefish_model.h5"  # op HuggingFace gewoon in de root neerzetten
-IMG_SIZE = 64  # jouw resize in Kaggle
-
-CLASS_NAMES = [
-    'House Mackerel',
-    'Black Sea Sprat',
-    'Sea Bass',
-    'Red Mullet',
-    'Trout',
-    'Striped Red Mullet',
-    'Shrimp',
-    'Gilt-Head Bream',
-    'Red Sea Bream'
-]
-
-
-# ====== FUNCTIES ======
-@st.cache_resource
-def load_model():
-    """Laad het Keras-model één keer en cache het."""
-    model = tf.keras.models.load_model(MODEL_PATH)
-    return model
-
-
-def preprocess_image(image: Image.Image) -> np.ndarray:
-    """Resize + normaliseer afbeelding naar hetzelfde formaat als training."""
-    image = image.convert("RGB")
-    image = image.resize((IMG_SIZE, IMG_SIZE))
-    arr = np.array(image).astype("float32") / 255.0
-    arr = np.expand_dims(arr, axis=0)  # shape: (1, 64, 64, 3)
-    return arr
-
-
-def predict_image(model, image: Image.Image):
-    """Voorspel klasse + probabilities voor één afbeelding."""
-    x = preprocess_image(image)
-    preds = model.predict(x)[0]  # shape: (9,)
-    pred_idx = int(np.argmax(preds))
-    pred_class = CLASS_NAMES[pred_idx]
-    pred_conf = float(preds[pred_idx])
-    return pred_class, pred_conf, preds
-
-
-# ====== STREAMLIT UI ======
-st.set_page_config(
-    page_title="Fish Classifier",
-    page_icon="🐟",
-)
-
-st.title("🐟 Large-Scale Fish Classifier")
-st.write(
-    """
-Upload een afbeelding van een vis uit de dataset 
-en het model voorspelt de soort.
-"""
-)
-
-# Model alvast laden (toont spinner)
-with st.spinner("Model laden..."):
-    model = load_model()
-
-uploaded_file = st.file_uploader(
-    "Upload een afbeelding", type=["jpg", "jpeg", "png"]
-)
-
-if uploaded_file is not None:
-    image = Image.open(uploaded_file)
-    st.image(image, caption="Geüploade afbeelding", use_column_width=True)
-
-    if st.button("Classify"):
-        with st.spinner("Bezig met voorspellen..."):
-            pred_class, pred_conf, preds = predict_image(model, image)
-
-        st.subheader("Voorspelling")
-        st.write(f"**{pred_class}** met **{pred_conf:.2%}** zekerheid.")
-
-        # Probabilities plotten als bar chart
-        prob_dict = {CLASS_NAMES[i]: float(preds[i]) for i in range(len(CLASS_NAMES))}
-        st.subheader("Class probabilities")
-        st.bar_chart(prob_dict)
-
-else:
-    st.info("➡️ Upload eerst een afbeelding (jpg/jpeg/png).")
+import streamlit as st
+import numpy as np
+from PIL import Image
+import onnxruntime as ort
+
+# ====== MODEL SETTINGS ======
+MODEL_PATH = "cnn_largefish_model.onnx"
+IMG_SIZE = 64  # jouw trainingsgrootte
+
+CLASS_NAMES = [
+    'House Mackerel',
+    'Black Sea Sprat',
+    'Sea Bass',
+    'Red Mullet',
+    'Trout',
+    'Striped Red Mullet',
+    'Shrimp',
+    'Gilt-Head Bream',
+    'Red Sea Bream'
+]
+
+@st.cache_resource
+def load_session():
+    session = ort.InferenceSession(
+        MODEL_PATH,
+        providers=["CPUExecutionProvider"]
+    )
+    input_name = session.get_inputs()[0].name
+    return session, input_name
+
+def preprocess_image(image: Image.Image) -> np.ndarray:
+    image = image.convert("RGB")
+    image = image.resize((IMG_SIZE, IMG_SIZE))
+    arr = np.array(image).astype("float32") / 255.0
+    arr = np.expand_dims(arr, axis=0)  
+    return arr
+
+def predict(image: Image.Image):
+    session, input_name = load_session()
+    x = preprocess_image(image)
+    preds = session.run(None, {input_name: x})[0][0]
+    pred_idx = int(np.argmax(preds))
+    pred_class = CLASS_NAMES[pred_idx]
+    pred_conf = float(preds[pred_idx])
+    return pred_class, pred_conf, preds
+
+st.set_page_config(page_title="Fish Classifier", page_icon="🐟")
+
+st.title("🐟 Large-Scale Fish Classifier (ONNX)")
+st.write("Upload een afbeelding en het model voorspelt de soort.")
+
+uploaded_file = st.file_uploader("Upload een afbeelding", type=["jpg", "jpeg", "png"])
+
+if uploaded_file is not None:
+    image = Image.open(uploaded_file)
+    st.image(image, caption="Geüploade afbeelding", use_column_width=True)
+
+    if st.button("Classify"):
+        with st.spinner("Bezig met voorspellen..."):
+            pred_class, pred_conf, preds = predict(image)
+
+        st.subheader("Voorspelling")
+        st.write(f"**{pred_class}** met **{pred_conf:.2%}** zekerheid.")
+
+        st.subheader("Class probabilities")
+        st.bar_chart({CLASS_NAMES[i]: float(preds[i]) for i in range(len(CLASS_NAMES))})
+else:
+    st.info("➡️ Upload eerst een afbeelding.")