app.py

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

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():
    """
    Laad het ONNX-model één keer in een ONNX Runtime sessie.
    """
    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:
    """
    Resize + normaliseer naar (1, 64, 64, 3) met waarden 0–1.
    """
    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)  # (1, 64, 64, 3)
    return arr


def predict(image: Image.Image):
    """
    Run één voorspelling via ONNX Runtime.
    """
    session, input_name = load_session()
    x = preprocess_image(image)

    # ONNX Runtime geeft een list terug; [0] is de output tensor
    preds = session.run(None, {input_name: x})[0][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="Large-Scale Fish Classifier", page_icon="🐟")

st.title("🐟 Large-Scale Fish Classifier")
st.write("Upload een afbeelding van een vis 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 (jpg/jpeg/png).")