app.py

from PIL import Image
import streamlit as st
import cv2
import numpy as np
import os
import tensorflow as tf

from IMVIP_Supplementary_Material.scripts import dfutils #methods used for DF-Net

DESCRIPTION = """# DF-Net
The Digital Forensics Network is designed and trained to detect and locate image manipulations. 
More information can be found in this [publication](https://zenodo.org/record/8214996)

#### Select example image or upload your own image:
"""

IMG_SIZE=256

tf.experimental.numpy.experimental_enable_numpy_behavior()
#np.warnings.filterwarnings('error', category=np.VisibleDeprecationWarning) 

model_M1 = None
model_M2 = None

def check_forgery_df(img):
    shape_original = img.shape
    img = cv2.resize(img, (IMG_SIZE,IMG_SIZE))
    x = np.expand_dims( img.astype('float32')/255., axis=0 )
    
    pred1 = model_M1.predict(x, verbose=0)
    pred2= model_M2.predict(x, verbose=0)   


#    # Ensure pred1 and pred2 are numpy arrays before proceeding
#    if isinstance(pred1, dict):
#        print("pred1 is dict!")
#        pred1 = pred1[next(iter(pred1))]
#    if isinstance(pred2, dict):
#        pred2 = pred2[next(iter(pred2))]

    
    pred = np.max([pred1,pred2], axis=0)   
    
    pred = dfutils.create_mask(pred)
    pred = pred.reshape(pred.shape[-3:-1])
    resized_image = cv2.resize(pred, (shape_original[1],shape_original[0]), interpolation=cv2.INTER_LINEAR)

    return resized_image


def evaluate(img):
    pre_t = check_forgery_df(img)
    st.image(pre_t, caption="White area indicates potential image manipulations.")

def start_evaluation(uploaded_file):    
    #load models
    model_path1 = "IMVIP_Supplementary_Material/models/model1/"
    model_path2 = "IMVIP_Supplementary_Material/models/model2/"
    
    #tfsm_layer1 = tf.keras.layers.TFSMLayer(model_path1, call_endpoint='serving_default')
    #tfsm_layer2 = tf.keras.layers.TFSMLayer(model_path2, call_endpoint='serving_default')
    #
    #input_shape = (256, 256, 3)
    #inputs = Input(shape=input_shape)

    ##create the model
    #outputs1 = tfsm_layer1(inputs)
    #model_M1 = Model(inputs, outputs1)

    #outputs2 = tfsm_layer2(inputs)
    #model_M2 = Model(inputs, outputs2)
    
    model_M1 = tf.keras.models.load_model("IMVIP_Supplementary_Material/models/model1/") #tf.keras.models.load_model("IMVIP_Supplementary_Material/models/model1/")
    model_M2 = tf.keras.models.load_model("IMVIP_Supplementary_Material/models/model2/")

    # Convert the file to an opencv image.
    file_bytes = np.asarray(bytearray(uploaded_file.read()), dtype=np.uint8)
    opencv_image = cv2.imdecode(file_bytes, 1)
    reversed_image = opencv_image[:, :, ::-1]
    st.image(reversed_image, caption="Input Image")
    evaluate(reversed_image)


st.markdown(DESCRIPTION)


img_path1 = "example_images/Sp_D_NRD_A_nat0095_art0058_0582"
img_path2 = "example_images/Sp_D_NRN_A_nat0083_arc0080_0445"
img_path3 = "example_images/Sp_D_NRN_A_ani0088_cha0044_0441"
image_paths = [img_path1+".jpg", img_path2+".jpg", img_path3+".jpg"]
gt_paths = [img_path1+"_gt.png", img_path2+"_gt.png", img_path3+"_gt.png"]
# Display images in a table format
img = None
for idx, image_path in enumerate(image_paths):
    cols = st.columns([2, 2, 2, 2])  # Define column widths
    
    # Place the button in the first column
    if cols[0].button(f"Select Image {idx+1}", key=idx):
        img = Image.open(image_path)
        
    # Place the image in the second column
    with cols[1]:
        st.image(image_path, use_column_width=True, caption="Example Image "+str(idx+1))

    # Place the ground truth in the third column
    with cols[2]:
        st.image(gt_paths[idx], use_column_width=True, caption="Ground Truth")

if img is not None:
    start_evaluation(img)

def reset_image_select():
    img = None




def start_evaluation(uploaded_file):    
    #load models
    model_path1 = "IMVIP_Supplementary_Material/models/model1/"
    model_path2 = "IMVIP_Supplementary_Material/models/model2/"
    
    #tfsm_layer1 = tf.keras.layers.TFSMLayer(model_path1, call_endpoint='serving_default')
    #tfsm_layer2 = tf.keras.layers.TFSMLayer(model_path2, call_endpoint='serving_default')
    #
    #input_shape = (256, 256, 3)
    #inputs = Input(shape=input_shape)

    ##create the model
    #outputs1 = tfsm_layer1(inputs)
    #model_M1 = Model(inputs, outputs1)

    #outputs2 = tfsm_layer2(inputs)
    #model_M2 = Model(inputs, outputs2)
    
    model_M1 = tf.keras.models.load_model("IMVIP_Supplementary_Material/models/model1/") #tf.keras.models.load_model("IMVIP_Supplementary_Material/models/model1/")
    model_M2 = tf.keras.models.load_model("IMVIP_Supplementary_Material/models/model2/")

    # Convert the file to an opencv image.
    file_bytes = np.asarray(bytearray(uploaded_file.read()), dtype=np.uint8)
    opencv_image = cv2.imdecode(file_bytes, 1)
    reversed_image = opencv_image[:, :, ::-1]
    st.image(reversed_image, caption="Input Image")
    evaluate(reversed_image)


uploaded_file= None
uploaded_file = st.file_uploader("Please upload an image", type=["jpeg", "jpg", "png"], on_change=reset_image_select)
if (uploaded_file is not None) and (img is None):
    start_evaluation(uploaded_file)