Initial Commit

.gitattributes

@@ -0,0 +1,35 @@
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tar filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text

IMVIP_Supplementary_Material/.gitattributes

@@ -0,0 +1,2 @@
+models/model1 filter=lfs diff=lfs merge=lfs -text
+models/model2 filter=lfs diff=lfs merge=lfs -text

IMVIP_Supplementary_Material/models/model1/keras_metadata.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e60097affd2751ea4540b0659522b06c2755d0ec4c8aede29a1b397d960b7949
+size 194745

IMVIP_Supplementary_Material/models/model1/saved_model.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7664979c02d7a1015b700e66b80291276f336e9bf16d8a3b1ca527cf4b91e5fb
+size 1800047

IMVIP_Supplementary_Material/models/model1/variables/.gitattributes

@@ -0,0 +1 @@
+variables.data-00000-of-00001 filter=lfs diff=lfs merge=lfs -text

IMVIP_Supplementary_Material/models/model1/variables/variables.data-00000-of-00001

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:46ec82afc5d7ab67e757b20fa5c4c16845881dab62516c6405dc87b1a381f58b
+size 62888221

IMVIP_Supplementary_Material/models/model2/keras_metadata.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4a38bbb31c5972308f17290c18665747ab537a3a63eb986d02eb212b6a41f6c5
+size 261437

IMVIP_Supplementary_Material/models/model2/saved_model.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ec722ab01d8860361c3912f811630983204c34f9d34d92fd0f70773418bd945d
+size 2068453

IMVIP_Supplementary_Material/models/model2/variables/.gitattributes

@@ -0,0 +1 @@
+variables.data-00000-of-00001 filter=lfs diff=lfs merge=lfs -text

IMVIP_Supplementary_Material/models/model2/variables/variables.data-00000-of-00001

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f8be65ab6ff1e7ebebbcc7257daee8373c43819a909dfda5df1a61b59a4b17ca
+size 21416427

IMVIP_Supplementary_Material/scripts/dfutils.py

@@ -0,0 +1,170 @@
+import matplotlib.pyplot as plt
+import numpy as np
+from sklearn.metrics import roc_auc_score, f1_score, jaccard_score, accuracy_score
+import tensorflow as tf
+
+
+
+# create prediction mask
+def create_mask(pred_mask):
+  if pred_mask.shape[-1] > 1:
+      pred_mask = tf.argmax(pred_mask, axis=-1)
+      pred_mask = pred_mask[..., tf.newaxis]
+  
+  return pred_mask[0]
+
+
+
+def metric_copy(premask, groundtruth):
+    seg_inv, gt_inv = np.logical_not(premask), np.logical_not(groundtruth)
+    true_pos = float(np.logical_and(premask, groundtruth).sum())  # float for division
+    true_neg = np.logical_and(seg_inv, gt_inv).sum()
+    false_pos = np.logical_and(premask, gt_inv).sum()
+    false_neg = np.logical_and(seg_inv, groundtruth).sum()
+    f1 = 2 * true_pos / (2 * true_pos + false_pos + false_neg + 1e-6)
+    cross = np.logical_and(premask, groundtruth)
+    union = np.logical_or(premask, groundtruth)
+    iou = np.sum(cross) / (np.sum(union) + 1e-6)
+    if np.sum(cross) + np.sum(union) == 0:
+        iou = 1
+    return f1, iou
+
+
+
+def show_prediction(img, pred):
+    print("max_pred = ", np.max(pred), " min_pred = ", np.min(pred))
+    plt.subplot(1,2,1)
+    plt.imshow(img)
+    plt.subplot(1,2,2)
+    plt.imshow(pred, cmap='gray') #, vmin=0, vmax=1)
+    plt.show()
+
+
+def show_predictions(dataset=None, num=1):
+  if dataset:
+    for image, mask in dataset.take(num):
+      pred_mask = model.predict(image)
+      display([image[0], mask[0], create_mask(pred_mask)])
+  else:
+    print(sample_image.shape)
+    print(sample_mask.shape)
+    display([sample_image, sample_mask,
+             create_mask(model.predict(sample_image[tf.newaxis, ...]))])
+
+
+
+def display(display_list, reverseRGB = True):
+  plt.figure(figsize=(4, 4))
+
+  title = ['Input Image', 'True Mask', 'Predicted Mask']
+
+  for i in range(len(display_list)):
+    plt.subplot(1, len(display_list), i+1)
+    plt.title(title[i])
+    if reverseRGB:
+        plt.imshow(tf.keras.utils.array_to_img(display_list[i][...,::-1]))
+    else:
+        plt.imshow(tf.keras.utils.array_to_img(display_list[i]))
+    plt.axis('off')
+  plt.show()
+
+
+def get_gt_and_osn_folders(folder):
+    folder_list = [folder]
+    folder_list.append(folder+"_Facebook")
+    folder_list.append(folder+"_Whatsapp")
+    folder_list.append(folder+"_Weibo")
+    folder_list.append(folder+"_Wechat")
+    gt_folder = folder + "_GT"
+    return gt_folder,folder_list
+
+def get_gt_and_osn_folder(folder, osn):
+    osn_folder = folder+osn
+    gt_folder = folder + "_GT"
+    return gt_folder,osn_folder
+
+
+# plots the image + prediction + ground truth
+def plot_img_pred_gt(img_path, pre_t, gt):
+    print("INPUT plot_img_pred_gt:")
+    print("  img_path: ", img_path)
+    #get image
+    img = cv2.imread(img_path)
+    #plot image, prediction and mask
+    plot_img_pred_gt_execute(img,pre_t, gt)
+
+    
+def plot_img_pred_gt_execute(img, pre_t, gt, DISCRETIZE_OUTPUT=True):
+    #print("plot_img_pred_gt_execute(): pre_t.max: ", np.max(pre_t))
+    #print("plot_img_pred_gt_execute(): pre_t.min: ", np.min(pre_t))
+    if DISCRETIZE_OUTPUT:
+        pre_t = pre_t.numpy()
+        pre_t[pre_t > 0.5] = 1.0
+        pre_t[pre_t <= 0.5] = 0.0
+    plt.subplots(1,3,figsize=(10,10))
+    plt.subplot(1,3,1)
+    plt.imshow(img[...,::-1])
+    plt.title("Original Image")
+    plt.subplot(1,3,2)
+    plt.imshow(pre_t, cmap='gray')
+    #plt.imshow(pre_t>0.5, cmap='gray')
+    plt.title("Prediction")
+    plt.subplot(1,3,3)
+    plt.imshow(gt, cmap='gray')
+    plt.title("Ground Truth")
+    plt.show()
+    
+    
+def mask_bigger_fifty_perc(mask):
+    mask_size = mask.size
+    #print("mask.shape: ", mask.shape)
+    #print("mask_size: ", mask_size)
+    nr_points_in_mask = mask_size - (mask == 0.).sum()
+    mask_cover_perc_of_img = nr_points_in_mask/mask_size
+    #print("mask_cover_perc_of_img: ", mask_cover_perc_of_img)
+    if mask_cover_perc_of_img>0.5:
+        return True
+    return False
+
+
+#evaluation for one image (auc roc, f1, iou)
+def eval_image(pre_t, gt, auc, f1, iou, acc):
+    #df_out("pre_t_in evalimage",pre_t,True)
+
+    pre = np.repeat(pre_t.numpy()[:,:,np.newaxis],3,2)
+    H, W, _ = pre.shape
+    Hg, Wg, C = gt.shape
+
+    if mask_bigger_fifty_perc(gt):
+        print("FLIP pre because mask > 50% of image")
+        pre = 1 - pre
+    
+    if H != Hg or W != Wg:
+        print("ERROR: values not matching:")
+        print(f'H: {H}, W: {W}, C: {C}')
+        print(f'Hg: {Hg}, Wg: {Wg}, C: {C}')
+        gt = cv2.resize(gt, (W, H))
+        gt[gt > 127] = 255
+        gt[gt <= 127] = 0
+            
+    if np.max(gt) != np.min(gt):    
+        auc.append(roc_auc_score((gt.reshape(H*W*C) / 255.).astype('int'), pre.reshape(H*W*C)))
+    else:
+        print("!!!!!!!!!!!!!! eval_image(): np.max(gt) != np.min(gt) !!!!!!!!!!!!")
+    pre[pre>0.5] = 1.0
+    pre[pre<=0.5] = 0.0
+    
+    #consider changing to: a, b = metric_copy(pre , gt > 127)
+    #a, b = metric_copy(pre , gt / 255.) #old
+    a, b = metric_copy(pre , gt)
+        
+    
+    pre_ = tf.reshape(pre, [-1])
+    gt_ = tf.reshape(gt / 255., [-1]).astype(tf.int32)
+    acc_tmp = accuracy_score(pre_, gt_)
+    acc.append(acc_tmp)
+          
+    f1.append(a)
+    iou.append(b)
+    #print('Evaluation: AUC: %5.4f, F1: %5.4f, IOU: %5.4f' % (np.mean(auc), np.mean(f1), np.mean(iou)))
+

README.md

@@ -0,0 +1,13 @@
+---
+title: Df Net
+emoji: ⚡
+colorFrom: gray
+colorTo: gray
+sdk: streamlit
+sdk_version: 1.25.0
+app_file: app.py
+pinned: false
+license: cc-by-nc-sa-4.0
+---
+
+Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

app.py

@@ -0,0 +1,62 @@
+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)
+"""
+
+IMG_SIZE=256
+
+tf.experimental.numpy.experimental_enable_numpy_behavior()
+#np.warnings.filterwarnings('error', category=np.VisibleDeprecationWarning) 
+
+
+
+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)    
+    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.")
+
+
+
+st.markdown(DESCRIPTION)
+
+uploaded_file = st.file_uploader("Please upload an image", type=["jpeg", "jpg", "png"])
+
+if uploaded_file is not None:
+    #load models
+    model_M1 = 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)
+
+
+

packages.txt

@@ -0,0 +1 @@
+python3-opencv

requirements.txt

@@ -0,0 +1,5 @@
+opencv-python
+tensorflow
+pillow
+scikit-learn
+scikit-image