Update to handle big data

.gitignore

@@ -136,3 +136,4 @@ mtcnn/
 /split_dataset
 /prepared_dataset
 /App/uploads
+archive.zip

01b-crop_faces_with_azure-vision-api.py

@@ -1,89 +0,0 @@
-import cv2
-import sys, os.path
-import json
-import http.client, urllib.request, urllib.parse, urllib.error, base64
-
-base_path = '.\\train_sample_videos\\'
-AZURE_COMPUTER_VISION_NAME = '----REPLACE-WITH-YOUR-SERVICE-NAME----'  # e.g. xxxxxxxxxx.cognitiveservices.azure.com
-AZURE_COMPUTER_VISION_API_KEY = '----REPLACE-WITH-YOUR-KEY----'
-
-def get_filename_only(file_path):
-    file_basename = os.path.basename(file_path)
-    filename_only = file_basename.split('.')[0]
-    return filename_only
-
-with open(os.path.join(base_path, 'metadata.json')) as metadata_json:
-    metadata = json.load(metadata_json)
-    print(len(metadata))
-
-for filename in metadata.keys():
-    tmp_path = os.path.join(base_path, get_filename_only(filename))
-    print('Processing Directory: ' + tmp_path)
-    frame_images = [x for x in os.listdir(tmp_path) if os.path.isfile(os.path.join(tmp_path, x))]
-    faces_path = os.path.join(tmp_path, 'faces')
-    print('Creating Directory: ' + faces_path)
-    os.makedirs(faces_path, exist_ok=True)
-    print('Cropping Faces from Images...')
-
-    for frame in frame_images:
-        print('Processing ', frame)
-        image = cv2.cvtColor(cv2.imread(os.path.join(tmp_path, frame)), cv2.COLOR_BGR2RGB)
-
-        # Open the binary file
-        with open(os.path.join(tmp_path, frame), 'rb') as file_contents:
-            img_data = file_contents.read()
-
-        ######### Azure Computer Vision API
-        headers = {
-            # Request headers
-            'Content-Type': 'application/octet-stream',
-            'Ocp-Apim-Subscription-Key': AZURE_COMPUTER_VISION_API_KEY,
-        }
-
-        params = urllib.parse.urlencode({
-            # Request parameters
-            'visualFeatures': 'Faces'
-        })
-
-        try:
-            conn = http.client.HTTPSConnection(AZURE_COMPUTER_VISION_NAME)
-            conn.request("POST", "/vision/v3.0/analyze?%s" % params, img_data, headers)
-            response = conn.getresponse().read()
-            data = json.loads(response.decode('utf-8'))
-            print(data)
-            conn.close()
-        except Exception as e:
-            print("[Errno {0}] {1}".format(e.errno, e.strerror))
-            continue
-        
-        print(data['faces'])
-        print('Face Detected: ', len(data['faces']))
-        count = 0
-
-        for result in data['faces']:
-            bounding_box = []
-            bounding_box.append(result['faceRectangle']['left'])
-            bounding_box.append(result['faceRectangle']['top'])
-            bounding_box.append(result['faceRectangle']['width'])
-            bounding_box.append(result['faceRectangle']['height'])
-            print(bounding_box)
-
-            margin_x = bounding_box[2] * 0.3  # 30% as the margin
-            margin_y = bounding_box[3] * 0.3  # 30% as the margin
-            x1 = int(bounding_box[0] - margin_x)
-            if x1 < 0:
-                x1 = 0
-            x2 = int(bounding_box[0] + bounding_box[2] + margin_x)
-            if x2 > image.shape[1]:
-                x2 = image.shape[1]
-            y1 = int(bounding_box[1] - margin_y)
-            if y1 < 0:
-                y1 = 0
-            y2 = int(bounding_box[1] + bounding_box[3] + margin_y)
-            if y2 > image.shape[0]:
-                y2 = image.shape[0]
-            print(x1, y1, x2, y2)
-            crop_image = image[y1:y2, x1:x2]
-            new_filename = '{}-{:02d}.png'.format(os.path.join(faces_path, get_filename_only(frame)), count)
-            count = count + 1
-            cv2.imwrite(new_filename, cv2.cvtColor(crop_image, cv2.COLOR_RGB2BGR))

02-prepare_fake_real_dataset.py

@@ -3,6 +3,9 @@ import os
 import shutil
 import numpy as np
 import splitfolders as split_folders
+from PIL import Image
+
+MIN_IMAGE_SIZE = 90  # minimum width and height in pixels
 
 base_path = '.\\train_sample_videos\\Deepfakes\\'
 dataset_path = '.\\prepared_dataset\\'
@@ -11,13 +14,36 @@ os.makedirs(dataset_path, exist_ok=True)
 
 tmp_fake_path = '.\\tmp_fake_faces'
 print('Creating Directory: ' + tmp_fake_path)
-os.makedirs(tmp_fake_path, exist_ok=True)
+if os.path.exists(tmp_fake_path):
+    shutil.rmtree(tmp_fake_path)
+os.makedirs(tmp_fake_path)
 
 def get_filename_only(file_path):
     file_basename = os.path.basename(file_path)
     filename_only = file_basename.split('.')[0]
     return filename_only
 
+def copy_large_faces(src_dir, dst_dir):
+    """Copy only images that are at least MIN_IMAGE_SIZE x MIN_IMAGE_SIZE."""
+    skipped = 0
+    copied = 0
+    for fname in os.listdir(src_dir):
+        src_file = os.path.join(src_dir, fname)
+        if not os.path.isfile(src_file):
+            continue
+        try:
+            with Image.open(src_file) as img:
+                w, h = img.size
+            if w >= MIN_IMAGE_SIZE and h >= MIN_IMAGE_SIZE:
+                shutil.copy2(src_file, os.path.join(dst_dir, fname))
+                copied += 1
+                print(f'Copied: {copied}')
+            else:
+                print(f'Skipped {fname}: {w}x{h}')
+                skipped += 1
+        except Exception:
+            skipped += 1
+   
 with open(os.path.join('.\\train_sample_videos\\csv\\', 'Deepfakes.csv'), newline='', encoding='utf-8') as csvfile:
     reader = csv.DictReader(csvfile)
     metadata = {}
@@ -27,11 +53,15 @@ with open(os.path.join('.\\train_sample_videos\\csv\\', 'Deepfakes.csv'), newlin
 
 real_path = os.path.join(dataset_path, 'real')
 print('Creating Directory: ' + real_path)
-os.makedirs(real_path, exist_ok=True)
+if os.path.exists(real_path):
+    shutil.rmtree(real_path)
+os.makedirs(real_path)
 
 fake_path = os.path.join(dataset_path, 'fake')
 print('Creating Directory: ' + fake_path)
-os.makedirs(fake_path, exist_ok=True)
+if os.path.exists(fake_path):
+    shutil.rmtree(fake_path)
+os.makedirs(fake_path)
 
 for filename, label in metadata.items():
     print(filename)
@@ -41,10 +71,10 @@ for filename, label in metadata.items():
     if os.path.exists(tmp_path):
         if label == 'REAL':
             print('Copying to :' + real_path)
-            shutil.copytree(tmp_path, real_path, dirs_exist_ok=True)
+            copy_large_faces(tmp_path, real_path)
         elif label == 'FAKE':
             print('Copying to :' + tmp_fake_path)
-            shutil.copytree(tmp_path, tmp_fake_path, dirs_exist_ok=True)
+            copy_large_faces(tmp_path, tmp_fake_path)
         else:
             print('Ignored..')
 
@@ -54,13 +84,10 @@ print('Total Number of Real faces: ', len(all_real_faces))
 all_fake_faces = [f for f in os.listdir(tmp_fake_path) if os.path.isfile(os.path.join(tmp_fake_path, f))]
 print('Total Number of Fake faces: ', len(all_fake_faces))
 
-random_faces = np.random.choice(all_fake_faces, len(all_real_faces), replace=False)
-for fname in random_faces:
-    src = os.path.join(tmp_fake_path, fname)
-    dst = os.path.join(fake_path, fname)
-    shutil.copyfile(src, dst)
+print('Copying filtered fake faces to: ' + fake_path)
+copy_large_faces(tmp_fake_path, fake_path)
 
-print('Down-sampling Done!')
+print('Copying all fake faces Done!')
 
 # Split into Train/ Val/ Test folders
 split_folders.ratio(dataset_path, output='split_dataset', seed=1377, ratio=(.8, .1, .1)) # default values