File size: 19,991 Bytes
16a5fbe
 
 
 
 
 
 
3d8b532
 
 
 
 
 
 
 
 
 
16a5fbe
 
3d8b532
 
 
16a5fbe
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3d8b532
 
16a5fbe
3d8b532
 
 
 
16a5fbe
 
3d8b532
16a5fbe
 
 
3d8b532
16a5fbe
 
 
 
 
 
 
3d8b532
 
16a5fbe
3d8b532
16a5fbe
3d8b532
 
 
16a5fbe
 
 
 
 
3d8b532
 
 
 
 
 
16a5fbe
3d8b532
 
 
 
16a5fbe
3d8b532
 
 
 
 
16a5fbe
3d8b532
 
 
 
 
16a5fbe
3d8b532
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
16a5fbe
3d8b532
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
16a5fbe
3d8b532
 
 
 
 
16a5fbe
3d8b532
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
16a5fbe
3d8b532
 
 
 
 
 
 
16a5fbe
3d8b532
 
 
 
 
 
 
 
 
 
 
 
16a5fbe
3d8b532
 
 
 
 
 
16a5fbe
3d8b532
16a5fbe
 
3d8b532
16a5fbe
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3d8b532
16a5fbe
 
3d8b532
 
16a5fbe
 
 
 
 
 
 
 
 
 
 
3d8b532
 
 
 
 
 
16a5fbe
3d8b532
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
16a5fbe
 
 
3d8b532
16a5fbe
3d8b532
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
16a5fbe
 
 
3d8b532
 
 
 
 
16a5fbe
 
 
3d8b532
 
 
 
16a5fbe
 
 
 
 
3d8b532
 
16a5fbe
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
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
import os
os.environ['TF_ENABLE_ONEDNN_OPTS'] = '0'  # Disable oneDNN to avoid numerical differences warning
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'   # Suppress TensorFlow logs except critical errors

import logging
logging.getLogger('tensorflow').setLevel(logging.ERROR)  # Further suppress TensorFlow warnings

import altair as alt
import numpy as np
import pandas as pd
import streamlit as st
import cv2
import torch
from transformers import AutoImageProcessor, AutoModelForImageClassification
from collections import deque
import tensorflow as tf
from tensorflow.keras.models import load_model
import tempfile
import time
import urllib.request
import shutil

# Cached model loading functions
@st.cache_resource
def load_cnn_model():
    try:
        model = load_model('cnn_model.h5')
        st.success("CNN model loaded successfully!")
        return model
    except Exception as e:
        st.error(f"Error loading CNN model: {e}")
        st.warning("Please make sure 'cnn_model.h5' is in the current directory.")
        return None

@st.cache_resource
def load_vit_components():
    image_processor = AutoImageProcessor.from_pretrained('Adieee5/deepfake-detection-f3net-cross', use_fast=True)
    model = AutoModelForImageClassification.from_pretrained('Adieee5/deepfake-detection-f3net-cross')
    return image_processor, model

@st.cache_resource
def load_face_net():
    model_file = "deploy.prototxt"
    weights_file = "res10_300x300_ssd_iter_140000.caffemodel"
    if os.path.exists(model_file) and os.path.exists(weights_file):
        return cv2.dnn.readNetFromCaffe(model_file, weights_file)
    return None

@st.cache_resource
def load_haar_cascade():
    cascade_path = cv2.data.haarcascades + 'haarcascade_frontalface_default.xml'
    if os.path.exists(cascade_path):
        return cv2.CascadeClassifier(cascade_path)
    return None

class CNNDeepfakeDetector:
    def __init__(self):
        self.model = load_cnn_model()

class DeepfakeDetector:
    def __init__(self):
        st.info("Initializing Deepfake Detector... This may take a moment.")
        
        # Load ViT components
        with st.spinner("Loading deepfake detection model..."):
            self.image_processor, self.model = load_vit_components()
        
        # Load face detection models
        with st.spinner("Loading face detection model..."):
            self.face_net = load_face_net()
            self.use_dnn = self.face_net is not None
            if self.use_dnn:
                st.success("Using DNN face detector (better for close-up faces)")
            else:
                self.face_cascade = load_haar_cascade()
                if self.face_cascade:
                    st.warning("Using Haar cascade face detector as fallback")
                else:
                    st.error(f"Cascade file not found")

        # Initialize CNN detector
        self.cnn_detector = CNNDeepfakeDetector()

        # Face tracking/smoothing parameters
        self.face_history = {}
        self.face_history_max_size = 10
        self.face_ttl = 5
        self.next_face_id = 0
        self.result_buffer_size = 5
        self.processing_times = deque(maxlen=30)

        st.success("Models loaded successfully!")

    def detect_faces_haar(self, frame):
        gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
        faces = self.face_cascade.detectMultiScale(gray, scaleFactor=1.1, minNeighbors=5, minSize=(30, 30))
        return [(x, y, w, h, 0.8) for (x, y, w, h) in faces]

    def detect_faces_dnn(self, frame):
        height, width = frame.shape[:2]
        blob = cv2.dnn.blobFromImage(cv2.resize(frame, (300, 300)), 1.0, (300, 300), (104.0, 177.0, 123.0))
        self.face_net.setInput(blob)
        detections = self.face_net.forward()
        faces = []
        for i in range(detections.shape[2]):
            confidence = detections[0, 0, i, 2]
            if confidence > 0.5:
                box = detections[0, 0, i, 3:7] * np.array([width, height, width, height])
                (x1, y1, x2, y2) = box.astype("int")
                x1, y1 = max(0, x1), max(0, y1)
                x2, y2 = min(width, x2), min(height, y2)
                w, h = x2 - x1, y2 - y1
                if w > 0 and h > 0:
                    faces.append((x1, y1, w, h, confidence))
        return faces

    def calculate_iou(self, box1, box2):
        box1_x1, box1_y1, box1_w, box1_h = box1
        box2_x1, box2_y1, box2_w, box2_h = box2
        box1_x2, box1_y2 = box1_x1 + box1_w, box1_y1 + box1_h
        box2_x2, box2_y2 = box2_x1 + box2_w, box2_y1 + box2_h
        x_left = max(box1_x1, box2_x1)
        y_top = max(box1_y1, box2_y1)
        x_right = min(box1_x2, box2_x2)
        y_bottom = min(box1_y2, box2_y2)
        if x_right < x_left or y_bottom < y_top:
            return 0.0
        intersection_area = (x_right - x_left) * (y_bottom - y_top)
        box1_area = box1_w * box1_h
        box2_area = box2_w * box2_h
        return intersection_area / float(box1_area + box2_area - intersection_area)

    def track_faces(self, faces):
        matched_faces = []
        unmatched_detections = list(range(len(faces)))
        if not self.face_history:
            for face in faces:
                face_id = self.next_face_id
                self.next_face_id += 1
                self.face_history[face_id] = {
                    'positions': deque([face[:4]], maxlen=self.face_history_max_size),
                    'ttl': self.face_ttl,
                    'label': None,
                    'confidence': 0.0,
                    'result_history': deque(maxlen=self.result_buffer_size)
                }
                matched_faces.append((face_id, face))
            return matched_faces

        for face_id in list(self.face_history.keys()):
            last_pos = self.face_history[face_id]['positions'][-1]
            best_match = -1
            best_iou = 0.3
            for i in unmatched_detections:
                iou = self.calculate_iou(last_pos, faces[i][:4])
                if iou > best_iou:
                    best_iou = iou
                    best_match = i
            if best_match != -1:
                matched_face = faces[best_match]
                self.face_history[face_id]['positions'].append(matched_face[:4])
                self.face_history[face_id]['ttl'] = self.face_ttl
                matched_faces.append((face_id, matched_face))
                unmatched_detections.remove(best_match)
            else:
                self.face_history[face_id]['ttl'] -= 1
                if self.face_history[face_id]['ttl'] <= 0:
                    del self.face_history[face_id]
                else:
                    predicted_face = (*last_pos, 0.5)
                    matched_faces.append((face_id, predicted_face))

        for i in unmatched_detections:
            face_id = self.next_face_id
            self.next_face_id += 1
            self.face_history[face_id] = {
                'positions': deque([faces[i][:4]], maxlen=self.face_history_max_size),
                'ttl': self.face_ttl,
                'label': None,
                'confidence': 0.0,
                'result_history': deque(maxlen=self.result_buffer_size)
            }
            matched_faces.append((face_id, faces[i]))
        return matched_faces

    def smooth_face_position(self, face_id):
        positions = self.face_history[face_id]['positions']
        if len(positions) == 1:
            return positions[0]
        total_weight = 0
        x, y, w, h = 0, 0, 0, 0
        for i, pos in enumerate(positions):
            weight = 2 ** i
            total_weight += weight
            x += pos[0] * weight
            y += pos[1] * weight
            w += pos[2] * weight
            h += pos[3] * weight
        return (int(x / total_weight), int(y / total_weight), int(w / total_weight), int(h / total_weight))

    def update_face_classification(self, face_id, label, confidence):
        self.face_history[face_id]['result_history'].append((label, confidence))
        real_votes = 0
        fake_votes = 0
        total_confidence = 0.0
        for result_label, result_conf in self.face_history[face_id]['result_history']:
            if result_label == "Real":
                real_votes += 1
                total_confidence += result_conf
            elif result_label == "Fake":
                fake_votes += 1
                total_confidence += result_conf
        if real_votes >= fake_votes:
            smoothed_label = "Real"
            label_confidence = real_votes / len(self.face_history[face_id]['result_history'])
        else:
            smoothed_label = "Fake"
            label_confidence = fake_votes / len(self.face_history[face_id]['result_history'])
        avg_confidence = (total_confidence / len(self.face_history[face_id]['result_history'])) * label_confidence
        self.face_history[face_id]['label'] = smoothed_label
        self.face_history[face_id]['confidence'] = avg_confidence
        return smoothed_label, avg_confidence

    def process_video(self, video_path, stframe, status_text, progress_bar, detector_type="dnn", model_type="vit"):
        use_dnn_current = detector_type == "dnn" and self.use_dnn
        cap = cv2.VideoCapture(video_path)
        if not cap.isOpened():
            st.error(f"Error: Cannot open video source")
            return
        frame_width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
        frame_height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
        fps = cap.get(cv2.CAP_PROP_FPS)
        total_frames = 250 if video_path != 0 else 0
        if video_path != 0:
            status_text.text(f"Video Info: {frame_width}x{frame_height}, {fps:.1f} FPS, {total_frames} frames")
        else:
            status_text.text(f"Webcam: {frame_width}x{frame_height}")
        self.face_history = {}
        self.next_face_id = 0
        self.processing_times = deque(maxlen=30)
        frame_count = 0
        process_every_n_frames = 2
        face_stats = {"Real": 0, "Fake": 0, "Unknown": 0}

        while True:
            start_time = time.time()
            ret, frame = cap.read()
            if not ret:
                status_text.text("End of video reached")
                break
            frame_count += 1
            if frame_count == 250:
                st.success("Video Processed Successfully!")
                break
            if video_path != 0:
                progress = min(float(frame_count) / float(max(total_frames, 1)), 1.0)
                progress_bar.progress(progress)
            process_frame = (frame_count % process_every_n_frames == 0)
            frame_rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)

            if process_frame:
                faces = self.detect_faces_dnn(frame) if use_dnn_current else self.detect_faces_haar(frame)
                tracked_faces = self.track_faces(faces)
                face_images = []
                face_ids = []
                for face_id, (x, y, w, h, face_confidence) in tracked_faces:
                    if face_id in self.face_history and w > 20 and h > 20:
                        sx, sy, sw, sh = self.smooth_face_position(face_id)
                        face = frame_rgb[sy:sy+sh, sx:sx+sw]
                        if face.size > 0 and face.shape[0] >= 20 and face.shape[1] >= 20:
                            face_images.append(face)
                            face_ids.append(face_id)
                if face_images:
                    if model_type == "vit":
                        inputs = self.image_processor(images=face_images, return_tensors="pt")
                        with torch.no_grad():
                            outputs = self.model(**inputs)
                        logits = outputs.logits
                        probs = torch.nn.functional.softmax(logits, dim=1)
                        preds = torch.argmax(logits, dim=1)
                        for i, pred in enumerate(preds):
                            label = 'Real' if pred.item() == 1 else 'Fake'
                            confidence = probs[i][pred].item()
                            self.update_face_classification(face_ids[i], label, confidence)
                    elif model_type == "cnn" and self.cnn_detector.model is not None:
                        img_arrays = [cv2.resize(face, (128, 128)) / 255.0 for face in face_images]
                        img_batch = np.array(img_arrays)
                        predictions = self.cnn_detector.model.predict(img_batch)
                        for i, prediction in enumerate(predictions):
                            confidence = float(prediction[0])
                            label = 'Real' if confidence < 0.5 else 'Fake'
                            if label == 'Fake':
                                confidence = confidence
                            else:
                                confidence = 1.0 - confidence
                            self.update_face_classification(face_ids[i], label, confidence)

            for face_id in self.face_history:
                if self.face_history[face_id]['ttl'] > 0:
                    sx, sy, sw, sh = self.smooth_face_position(face_id)
                    cv2.rectangle(frame, (sx, sy), (sx+sw, sy+sh), (0, 255, 255), 2)
                    label = self.face_history[face_id]['label'] or "Unknown"
                    confidence = self.face_history[face_id]['confidence']
                    result_text = f"{label}: {confidence:.2f}"
                    text_color = (0, 255, 0) if label == "Real" else (0, 0, 255)
                    cv2.rectangle(frame, (sx, sy+sh), (sx+len(result_text)*11, sy+sh+25), (0, 0, 0), -1)
                    cv2.putText(frame, result_text, (sx, sy+sh+20),
                                cv2.FONT_HERSHEY_SIMPLEX, 0.7, text_color, 2)
                    cv2.putText(frame, f"ID:{face_id}", (sx, sy-5),
                                cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 255, 0), 1)
                    if label in face_stats:
                        face_stats[label] += 1

            process_time = time.time() - start_time
            self.processing_times.append(process_time)
            avg_time = sum(self.processing_times) / len(self.processing_times)
            effective_fps = 1.0 / avg_time if avg_time > 0 else 0

            if video_path != 0:
                progress_percent = (frame_count / total_frames) * 100 if total_frames > 0 else 0
                cv2.putText(frame, f"Frame: {frame_count}/{total_frames} ({progress_percent:.1f}%)",
                           (10, 30), cv2.FONT_HERSHEY_SIMPLEX, 0.7, (255, 255, 255), 2)
            else:
                cv2.putText(frame, f"Frame: {frame_count}",
                           (10, 30), cv2.FONT_HERSHEY_SIMPLEX, 0.7, (255, 255, 255), 2)
            detector_name = "DNN" if use_dnn_current else "Haar Cascade"
            model_name = "ViT" if model_type == "vit" else "CNN"
            cv2.putText(frame, f"Detector: {detector_name} | Model: {model_name} | FPS: {effective_fps:.1f}",
                       (10, 60), cv2.FONT_HERSHEY_SIMPLEX, 0.6, (255, 255, 255), 2)
            cv2.putText(frame, f"Tracked faces: {len(self.face_history)}",
                       (10, 90), cv2.FONT_HERSHEY_SIMPLEX, 0.6, (255, 255, 255), 2)
            stframe.image(cv2.cvtColor(frame, cv2.COLOR_BGR2RGB), channels="RGB")
            status_text.text(f"Real: {face_stats['Real']} | Fake: {face_stats['Fake']} | FPS: {effective_fps:.1f}")
            if st.session_state.get('stop_button', False):
                break
        cap.release()
        return face_stats

def ensure_sample_video():
    sample_dir = "sample_videos"
    sample_path = os.path.join(sample_dir, "Sample.mp4")
    if not os.path.exists(sample_dir):
        os.makedirs(sample_dir)
    if not os.path.exists(sample_path):
        try:
            with st.spinner("Downloading sample video..."):
                sample_url = "https://storage.googleapis.com/deepfake-demo/sample_deepfake.mp4"
                with urllib.request.urlopen(sample_url) as response, open(sample_path, 'wb') as out_file:
                    shutil.copyfileobj(response, out_file)
                st.success("Sample video downloaded successfully!")
        except Exception as e:
            st.error(f"Failed to download sample video: {e}")
            return None
    return sample_path

def main():
    st.set_page_config(page_title="Deepfake Detector", layout="wide")
    st.title("Deepfake Detection App")
    st.markdown("""
    This app uses computer vision and deep learning to detect deepfake videos.
    Upload a video or use your webcam to detect if faces are real or manipulated.
    """)

    if 'detector' not in st.session_state:
        st.session_state.detector = None
    if 'stop_button' not in st.session_state:
        st.session_state.stop_button = False
    if 'use_sample' not in st.session_state:
        st.session_state.use_sample = False
    if 'sample_path' not in st.session_state:
        st.session_state.sample_path = None

    if st.session_state.detector is None:
        st.session_state.detector = DeepfakeDetector()

    st.sidebar.title("Options")
    input_option = st.sidebar.radio("Select Input Source", ["Upload Video", "Use Webcam", "Try Sample Video"])
    detector_type = st.sidebar.selectbox("Face Detector", ["DNN (better for close-ups)", "Haar Cascade (faster)"],
                                         index=0 if st.session_state.detector.use_dnn else 1)
    detector_option = "dnn" if "DNN" in detector_type else "haar"
    model_type = st.sidebar.selectbox("Deepfake Detection Model", ["Vision Transformer (ViT)", "F3 Net Model"], index=0)
    model_option = "vit" if "Vision" in model_type else "cnn"

    col1, col2 = st.columns([3, 1])
    with col1:
        video_placeholder = st.empty()
    with col2:
        status_text = st.empty()
        progress_bar = st.empty()
        st.subheader("Results")
        results_area = st.empty()
        if st.button("Stop Processing"):
            st.session_state.stop_button = True

    if input_option == "Upload Video":
        uploaded_file = st.sidebar.file_uploader("Choose a video file", type=["mp4", "avi", "mov", "mkv"])
        if uploaded_file is not None:
            st.session_state.stop_button = False
            tfile = tempfile.NamedTemporaryFile(delete=False)
            tfile.write(uploaded_file.read())
            video_path = tfile.name
            face_stats = st.session_state.detector.process_video(video_path, video_placeholder, status_text,
                                                                 progress_bar, detector_option, model_option)
            results_df = {"Category": ["Real Faces", "Fake Faces"], "Count": [face_stats["Real"], face_stats["Fake"]]}
            results_area.dataframe(results_df)
            os.unlink(video_path)
    elif input_option == "Use Webcam":
        st.session_state.stop_button = False
        if st.sidebar.button("Start Webcam"):
            face_stats = st.session_state.detector.process_video(0, video_placeholder, status_text, progress_bar,
                                                                 detector_option, model_option)
            results_df = {"Category": ["Real Faces", "Fake Faces"], "Count": [face_stats["Real"], face_stats["Fake"]]}
            results_area.dataframe(results_df)
    elif input_option == "Try Sample Video":
        st.session_state.stop_button = False
        sample_path = ensure_sample_video()
        if sample_path and st.sidebar.button("Process Sample Video"):
            face_stats = st.session_state.detector.process_video(sample_path, video_placeholder, status_text,
                                                                 progress_bar, detector_option, model_option)
            results_df = {"Category": ["Real Faces", "Fake Faces"], "Count": [face_stats["Real"], face_stats["Fake"]]}
            results_area.dataframe(results_df)

if __name__ == "__main__":
    main()