app.py

import streamlit as st
import torch
import torch.nn as nn
from PIL import Image
import albumentations as A
from albumentations.pytorch import ToTensorV2
import timm
import numpy as np
import onnxruntime as ort

import warnings
warnings.filterwarnings("ignore")

# Khởi tạo mô hình
class SiameseNetwork(nn.Module):
    def __init__(self, model_name='resnet18', pretrained=True):
        super(SiameseNetwork, self).__init__()
        self.encoder = timm.create_model(model_name, pretrained=pretrained, num_classes=0)

        self.fc = nn.Sequential(
            nn.Linear(self.encoder.num_features, 256),
            nn.ReLU(inplace=True),
            nn.Linear(256, 128)
        )

    def forward_once(self, x):
        output = self.encoder(x)
        output = self.fc(output)
        return output

    def forward(self, img1, img2):
        output1 = self.forward_once(img1)
        output2 = self.forward_once(img2)
        return output1, output2

# Load the model
@st.cache_resource
def load_model():
    backbone = 'efficientnet_b2'
    onnx_file_path = f"weights/model_{backbone}.onnx"
    session = ort.InferenceSession(onnx_file_path)
    return session

# Function to run inference
def inference_with_real_set(session, img, real_imgs, threshold=0.5):
    distances = []
    for real_img in real_imgs:
        input_dict = {
            session.get_inputs()[0].name: img.numpy(),
            session.get_inputs()[1].name: real_img.numpy()
        }
        outputs = session.run(None, input_dict)
        
        euclidean_distance = np.linalg.norm(outputs[0] - outputs[1])
        
        distances.append(euclidean_distance)

    avg_distance = sum(distances) / len(distances)
    
    return "Real" if avg_distance < threshold else "Fake"

# Image preprocessing function
def preprocess_image(image):
    transform = A.Compose([
        A.Resize(256, 256),
        A.Normalize(mean=(0.5, 0.5, 0.5), std=(0.5, 0.5, 0.5)),
        ToTensorV2(),
    ])
    image = np.array(image)/255.0
    image = transform(image=image)['image'].unsqueeze(0)
    return image

# Hide Streamlit warnings and footers
hide_warning = """
    <style>
    .stAlert {display: none;}
    .css-18e3th9 {padding-top: 2rem; padding-bottom: 2rem;}  /* Adjust top and bottom padding */
    .css-1d391kg {max-width: 100% !important; padding-left: 1rem; padding-right: 1rem;}  /* Adjust the max-width and side padding */
    </style>
    """
    
st.markdown(hide_warning, unsafe_allow_html=True)

# Streamlit interface
st.title("Forge Signature Siamese")

# Load the model
model = load_model()

# Chia màn hình thành 2 cột
col1, col2 = st.columns([1, 1])  # Cột bên trái để input, cột bên phải để hiển thị kết quả

with col1:
    # Upload image to compare
    uploaded_image = st.file_uploader("Upload an image to verify", type=["png", "jpg", "jpeg"])

    # Upload real reference images
    uploaded_real_images = st.file_uploader("Upload real reference images", type=["png", "jpg", "jpeg"], accept_multiple_files=True)

    # Threshold slider
    threshold = st.slider("Threshold", 0.0, 1.0, 0.5)

# Nếu cả hình ảnh tải lên và hình ảnh thật đã được tải lên
if uploaded_image is not None and uploaded_real_images:
    # Xử lý ảnh
    img = preprocess_image(Image.open(uploaded_image).convert('RGB'))
    real_imgs = [preprocess_image(Image.open(img).convert('RGB')) for img in uploaded_real_images]

    # Chạy inference
    result = inference_with_real_set(model, img, real_imgs, threshold)

    with col2:
        # Hiển thị nút kết quả với màu sắc tùy chỉnh
        if result == "Real":
            button_color = "background-color: green; color: white; font-weight: bold;"
            button_label = "Real"
        else:
            button_color = "background-color: red; color: white; font-weight: bold;"
            button_label = "Fake"
        
        st.markdown(
            f"""
            <div style="display: flex; justify-content: center; margin-top: 20px;">
                <button style="padding: 10px 20px; font-size: 18px; {button_color}">{button_label}</button>
            </div>
            """, 
            unsafe_allow_html=True
        )
        
        # Hiển thị hình ảnh và tham chiếu
        sub_col1, sub_col2 = st.columns([1, 1])

        # Cột nhỏ bên trái: Hình ảnh tải lên
        with sub_col1:
            st.write("\n")
            st.write("**Input Image**")
            st.image(uploaded_image, caption="Uploaded Image", use_column_width=True)

        # Cột nhỏ bên phải: Hình ảnh tham chiếu
        with sub_col2:
            st.write("\n")
            st.write("**Real Reference Images**")
            for real_image in uploaded_real_images:
                st.image(real_image, caption="Real Image", use_column_width=True)