wabala69/EATC-Assignment

1import streamlit as st2from tensorflow.keras.models import load_model3from tensorflow.keras.preprocessing import image4import numpy as np5from PIL import Image6 7# Load trained model8model = load_model("src/cnn_model.h5")9 10# Map class indices11class_indices = {'FAKE': 0, 'REAL': 1}12labels = {v: k for k, v in class_indices.items()}13 14# Image size (must match your model's input)15IMG_HEIGHT = 25616IMG_WIDTH = 25617 18# Streamlit UI19st.title("🕵️‍♂️ Deepfake Image Detector")20st.write("Upload an image and this app will tell you whether it is likely a **deepfake** or **real**.")21 22uploaded_file = st.file_uploader("Choose an image...", type=["jpg", "jpeg", "png"])23 24if uploaded_file is not None:25    # Display uploaded image26    img = Image.open(uploaded_file)27    st.image(img, caption="Uploaded Image", use_column_width=True)28 29    # Preprocess image30    img = img.convert('RGB')              # ensure 3 channels31    img = img.resize((IMG_WIDTH, IMG_HEIGHT))32    img_array = image.img_to_array(img)33    img_array = img_array / 255.034    img_array = np.expand_dims(img_array, axis=0)35 36    # Predict37    prediction = model.predict(img_array)[0][0]38    predicted_class = int(np.round(prediction))39    confidence = prediction if predicted_class == 1 else 1 - prediction40 41    # Output42    st.markdown("---")43    st.subheader("🔍 Prediction:")44    st.write(f"**Class:** {labels[predicted_class]}")45    st.write(f"**Confidence:** {confidence * 100:.2f}%")46    st.write(f"Raw prediction: {prediction}")47    st.write(f"**Predicted class:** {predicted_class} | **Raw score:** {prediction:.4f} | **Mapped label:** {labels[predicted_class]}")48 49