File size: 8,256 Bytes
2cfb6b0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a6b0e12
 
 
 
2cfb6b0
a6b0e12
 
8c6a131
 
 
 
 
 
 
28eea81
 
 
 
 
 
 
 
 
 
 
 
8c6a131
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a6b0e12
2cfb6b0
8c6a131
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a6b0e12
 
e5a0b5e
 
8c6a131
a6b0e12
e5a0b5e
 
 
8c6a131
 
a6b0e12
 
 
 
 
 
 
8c6a131
a6b0e12
 
 
8c6a131
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
# import gradio as gr
# import cv2
# import torch
# import dlib
# import numpy as np
# from imutils import face_utils
# from torchvision import models, transforms
# from tempfile import NamedTemporaryFile
# import shutil
# # Load face detector and landmark predictor
# face_detector = dlib.get_frontal_face_detector()
# PREDICTOR_PATH = "./shape_predictor_81_face_landmarks.dat"
# face_predictor = dlib.shape_predictor(PREDICTOR_PATH)

# import torch
# import torchvision.models as models

# # Load pretrained ResNet-34 model
# resnet34 = models.resnet34(weights=models.ResNet34_Weights.IMAGENET1K_V1)
# resnet34.fc = torch.nn.Linear(resnet34.fc.in_features, 2)
# ckpt_path = "./resnet34.pkl"

# # Save model state dict
# torch.save(resnet34.state_dict(), ckpt_path)
# print(f"✅ Model saved at {ckpt_path}")

# # Load deepfake detection model
# model = models.resnet34()
# model.fc = torch.nn.Linear(model.fc.in_features, 2)
# model.load_state_dict(torch.load(ckpt_path, map_location="cpu"))
# model.eval()

# # Define transformation for face images
# transform = transforms.Compose([
#     transforms.ToPILImage(),
#     transforms.Resize((224, 224)),
#     transforms.ToTensor(),
#     transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
# ])

# def process_video(video_path: str):
#     cap = cv2.VideoCapture(video_path)
#     width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
#     height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
#     fps = int(cap.get(cv2.CAP_PROP_FPS))

#     output_path = video_path.replace(".mp4", "_processed.mp4")
#     output_video = cv2.VideoWriter(output_path, cv2.VideoWriter_fourcc(*'mp4v'), fps, (width, height))

#     while cap.isOpened():
#         ret, frame = cap.read()
#         if not ret:
#             break

#         rgb_frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
#         faces = face_detector(rgb_frame, 1)

#         for face in faces:
#             landmarks = face_utils.shape_to_np(face_predictor(rgb_frame, face))
#             x_min, y_min = np.min(landmarks, axis=0)
#             x_max, y_max = np.max(landmarks, axis=0)

#             face_crop = rgb_frame[y_min:y_max, x_min:x_max]
#             if face_crop.size == 0:
#                 continue

#             face_tensor = transform(face_crop).unsqueeze(0)
#             with torch.no_grad():
#                 output = torch.softmax(model(face_tensor), dim=1)
#                 fake_confidence = output[0, 1].item() * 100  # Fake confidence as a percentage
#                 label = "Fake" if fake_confidence > 50 else "Real"
#                 color = (0, 0, 255) if label == "Fake" else (0, 255, 0)

#                 # Annotating confidence score with label
#                 label_text = f"{label} ({fake_confidence:.2f}%)"

#             cv2.rectangle(frame, (x_min, y_min), (x_max, y_max), color, 2)
#             cv2.putText(frame, label_text, (x_min, y_min - 10), cv2.FONT_HERSHEY_SIMPLEX, 0.5, color, 1)

#         output_video.write(frame)

#     cap.release()
#     output_video.release()
#     return output_path

# def gradio_interface(video_file):
#     if video_file is None:
#         return "Error: No video uploaded."

#     # Create a temporary file and copy the uploaded video content
#     with NamedTemporaryFile(delete=False, suffix=".mp4") as temp_file:
#         temp_file_path = temp_file.name
#         # Read the uploaded video file using its path
#         with open(video_file, "rb") as uploaded_file:
#             temp_file.write(uploaded_file.read())

#     output_path = process_video(temp_file_path)
#     return output_path

# # Gradio UI
# iface = gr.Interface(
#     fn=gradio_interface,
#     inputs=gr.Video(label="Upload Video"),
#     outputs=gr.Video(label="Processed Video"),
#     title="Deepfake Detection",
#     description="Upload a video to detect deepfakes. The model will process faces and classify them as real or fake."
# )

# if __name__ == "__main__":
#     iface.launch()









import gradio as gr
import cv2
import torch
import numpy as np
import mediapipe as mp
from torchvision import models, transforms
from tempfile import NamedTemporaryFile

# Initialize MediaPipe Face Detection and Face Mesh
mp_face_detection = mp.solutions.face_detection
mp_face_mesh = mp.solutions.face_mesh
face_detection = mp_face_detection.FaceDetection(model_selection=1, min_detection_confidence=0.5)
face_mesh = mp_face_mesh.FaceMesh(static_image_mode=False, max_num_faces=1, min_detection_confidence=0.5)

def create_model():
    model = models.resnet34(pretrained=False)  # Must match the architecture used during training
    model.fc = torch.nn.Linear(model.fc.in_features, 2)  # Adjust final layer
    return model

# Load the trained model
model = create_model()
try:
    model.load_state_dict(torch.load("resnet34.pkl", map_location='cpu'))
except RuntimeError as e:
    print(f"Error loading model: {e}")
    # Handle architecture mismatch (e.g., load weights manually)
model.eval()

# Define transformation for face images
transform = transforms.Compose([
    transforms.ToPILImage(),
    transforms.Resize((224, 224)),
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
])

def get_face_bbox(landmarks, frame_shape):
    h, w = frame_shape[:2]
    xs = [lm.x * w for lm in landmarks.landmark]
    ys = [lm.y * h for lm in landmarks.landmark]
    return int(min(xs)), int(min(ys)), int(max(xs)), int(max(ys))

def process_video(video_path: str):
    cap = cv2.VideoCapture(video_path)
    width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
    height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
    fps = int(cap.get(cv2.CAP_PROP_FPS))

    output_path = video_path.replace(".mp4", "_processed.mp4")
    output_video = cv2.VideoWriter(output_path, cv2.VideoWriter_fourcc(*'mp4v'), fps, (width, height))

    while cap.isOpened():
        ret, frame = cap.read()
        if not ret:
            break

        rgb_frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
        
        # Face detection
        results = face_detection.process(rgb_frame)
        if results.detections:
            for detection in results.detections:
                # Get face landmarks
                mesh_results = face_mesh.process(rgb_frame)
                if mesh_results.multi_face_landmarks:
                    for face_landmarks in mesh_results.multi_face_landmarks:
                        x_min, y_min, x_max, y_max = get_face_bbox(face_landmarks, frame.shape)
                        
                        face_crop = rgb_frame[y_min:y_max, x_min:x_max]
                        if face_crop.size == 0:
                            continue

                        face_tensor = transform(face_crop).unsqueeze(0)
                        with torch.no_grad():
                            output = torch.softmax(model(face_tensor), dim=1)
                            fake_confidence = output[0, 1].item() * 100
                            label = "Fake" if fake_confidence > 50 else "Real"
                            color = (0, 0, 255) if label == "Fake" else (0, 255, 0)
                            label_text = f"{label} ({fake_confidence:.2f}%)"

                        cv2.rectangle(frame, (x_min, y_min), (x_max, y_max), color, 2)
                        cv2.putText(frame, label_text, (x_min, y_min - 10), 
                                  cv2.FONT_HERSHEY_SIMPLEX, 0.5, color, 1)

        output_video.write(frame)

    cap.release()
    output_video.release()
    return output_path

def gradio_interface(video_file):
    if video_file is None:
        return "Error: No video uploaded."

    with NamedTemporaryFile(delete=False, suffix=".mp4") as temp_file:
        temp_file_path = temp_file.name
        with open(video_file, "rb") as uploaded_file:
            temp_file.write(uploaded_file.read())

    output_path = process_video(temp_file_path)
    return output_path

iface = gr.Interface(
    fn=gradio_interface,
    inputs=gr.Video(label="Upload Video"),
    outputs=gr.Video(label="Processed Video"),
    title="Deepfake Detection",
    description="Upload a video to detect deepfakes using MediaPipe face detection and ResNet-34 model."
)

if __name__ == "__main__":
    iface.launch(share=True)