app.py

import keras
import numpy as np
import streamlit as st
import random
import os
from PIL import Image, ImageOps
from io import BytesIO
from huggingface_hub import snapshot_download

def random_crop(img, min_size=160, max_size=2048, ratio=5/8):
    width, height = img.size

    crop_width = random.randint(min_size, min(max_size, width))
    crop_height = int(crop_width * ratio)

    if crop_height > height:
        crop_height = height
        crop_width = int(crop_height / ratio)
    
    left = random.randint(0, width - crop_width)
    top = random.randint(0, height - crop_height)
    right = left + crop_width
    bottom = top + crop_height

    return img.crop((left, top, right, bottom))

def jpg_compression(img):
    quality = random.randint(65, 100)
    jpeg_image = BytesIO()
    img.convert("RGB").save(jpeg_image, 'JPEG', quality=quality)
    jpeg_image.seek(0)
    compressed_img = Image.open(jpeg_image)
    return compressed_img
    
def get_prediction(img):
    x = np.array(img)
    x = np.expand_dims(x, axis=0)
    predictions = model.predict(x)
    return predictions[0,:]

models = ['DDPM', 'Glide', 'Latent Diffusion', 'Palette', 'Stable Diffusion', 'VQ Diffusion', 'real', 'unseen_fake']


st.title("Fake Detection")
st.divider()

st.subheader("Modelvariant")
variant = st.selectbox(
    "Choose the model",
    ("ResNet50v2-Basemodel", "ResNet50v2-Finetuned"),
    index=None,
    placeholder="Choose a model",
    label_visibility="hidden"
)
st.write("You selected model: ", variant)
binary = st.toggle("Activate binary classification")
st.divider()

if variant == "ResNet50v2-Basemodel":
    local_model_path = snapshot_download(repo_id="ElBeh/ma_basemodel")
else:
    local_model_path = snapshot_download(repo_id="ElBeh/ma_finetuned_model")


model = keras.models.load_model(local_model_path)

st.subheader("Image Preprocessing")
crop = st.toggle("random crop")
compress = st.toggle("jpeg compression")
st.divider()

file_name = st.file_uploader("Choose an image...")

#st.button("execute classification", type="primary")


if file_name is not None:
    col1, col2 = st.columns(2)
    image = Image.open(file_name)
    image = ImageOps.exif_transpose(image)
    
    if image.size != (200, 200) or image.mode != 'RGB':
        if crop:
            image = random_crop(image)
        image = image.resize((200, 200), Image.LANCZOS)     

    #if image.format != "JPEG" and compress :
    if  compress :
        image = jpg_compression(image)
        
    col1.image(image, use_column_width=True)  
    predictions = get_prediction(image)
    
    if binary:
        col2.header("Prediction")
        if predictions[6] > 0.5:
            col2.markdown(":green[real image!]")
        else:
            col2.markdown(":red[fake image!]")
    else:
        col2.header("Probabilities")
        if crop:
            st.button("re-crop")
        for idx,p in enumerate(predictions):
            col2.text(f"{ models[idx] }: { round(p * 100, 2)}%")