Create app.py

app.py

@@ -0,0 +1,328 @@
+import os
+import cv2
+import torch
+import numpy as np
+from PIL import Image
+import gradio as gr
+from gradio_client import Client, handle_file
+from torchvision.transforms import Normalize
+from facenet_pytorch.models.mtcnn import MTCNN
+import concurrent.futures
+import tempfile
+
+# ==========================================
+# 1. API ROUTER CONFIGURATION
+# ==========================================
+# These must match your exact Hugging Face Worker Space names
+WORKER_SPACES = [
+    "bithal26/DeepFake-Worker-1",
+    "bithal26/DeepFake-Worker-2",
+    "bithal26/DeepFake-Worker-3",
+    "bithal26/DeepFake-Worker-4",
+    "bithal26/DeepFake-Worker-5",
+    "bithal26/DeepFake-Worker-6",
+    "bithal26/DeepFake-Worker-7"
+]
+
+# Note: If your worker spaces are PRIVATE, you must add your HF_TOKEN 
+# to this UI Space's Secrets for the Client to connect successfully.
+clients = []
+print("Initializing connections to 7 API Workers...")
+for space in WORKER_SPACES:
+    try:
+        clients.append(Client(space))
+    except Exception as e:
+        print(f"Warning: Could not connect to {space}. Is it private/sleeping? Error: {e}")
+
+# ==========================================
+# 2. MTCNN PREPROCESSING ENGINE
+# ==========================================
+mean = [0.485, 0.456, 0.406]
+std = [0.229, 0.224, 0.225]
+normalize_transform = Normalize(mean, std)
+device = torch.device('cpu')
+
+def isotropically_resize_image(img, size, interpolation_down=cv2.INTER_AREA, interpolation_up=cv2.INTER_CUBIC):
+    h, w = img.shape[:2]
+    if max(w, h) == size: return img
+    scale = size / w if w > h else size / h
+    w, h = w * scale, h * scale
+    interpolation = interpolation_up if scale > 1 else interpolation_down
+    return cv2.resize(img, (int(w), int(h)), interpolation=interpolation)
+
+def put_to_center(img, input_size):
+    img = img[:input_size, :input_size]
+    image = np.zeros((input_size, input_size, 3), dtype=np.uint8)
+    start_w = (input_size - img.shape[1]) // 2
+    start_h = (input_size - img.shape[0]) // 2
+    image[start_h:start_h + img.shape[0], start_w: start_w + img.shape[1], :] = img
+    return image
+
+class VideoReader:
+    def read_frames(self, path, num_frames):
+        capture = cv2.VideoCapture(path)
+        frame_count = int(capture.get(cv2.CAP_PROP_FRAME_COUNT))
+        if frame_count <= 0: return None
+        frame_idxs = np.linspace(0, frame_count - 1, num_frames, endpoint=True, dtype=np.int32)
+        
+        frames, idxs_read = [], []
+        for frame_idx in range(frame_idxs[0], frame_idxs[-1] + 1):
+            ret = capture.grab()
+            if not ret: break
+            current = len(idxs_read)
+            if frame_idx == frame_idxs[current]:
+                ret, frame = capture.retrieve()
+                if not ret or frame is None: break
+                frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
+                frames.append(frame)
+                idxs_read.append(frame_idx)
+        capture.release()
+        return np.stack(frames), idxs_read if len(frames) > 0 else None
+
+class FaceExtractor:
+    def __init__(self):
+        self.video_reader = VideoReader()
+        self.detector = MTCNN(margin=0, thresholds=[0.7, 0.8, 0.8], device=device)
+
+    def process_video(self, video_path, frames_per_video=16): 
+        result = self.video_reader.read_frames(video_path, num_frames=frames_per_video)
+        if result is None: return []
+        my_frames, my_idxs = result
+        results = []
+        for frame in my_frames:
+            img = Image.fromarray(frame.astype(np.uint8))
+            img = img.resize(size=[s // 2 for s in img.size])
+            batch_boxes, probs = self.detector.detect(img, landmarks=False)
+            faces = []
+            if batch_boxes is not None:
+                for bbox in batch_boxes:
+                    if bbox is not None:
+                        xmin, ymin, xmax, ymax = [int(b * 2) for b in bbox]
+                        w, h = xmax - xmin, ymax - ymin
+                        p_h, p_w = h // 3, w // 3
+                        crop = frame[max(ymin - p_h, 0):ymax + p_h, max(xmin - p_w, 0):xmax + p_w]
+                        faces.append(crop)
+            if faces:
+                results.append({"faces": faces})
+        return results
+
+face_extractor = FaceExtractor()
+
+def confident_strategy(pred, t=0.8):
+    pred = np.array(pred)
+    sz = len(pred)
+    if sz == 0: return 0.0
+    fakes = np.count_nonzero(pred > t)
+    if fakes > sz // 2.5 and fakes > 11:
+        return np.mean(pred[pred > t])
+    elif np.count_nonzero(pred < 0.2) > 0.9 * sz:
+        return np.mean(pred[pred < 0.2])
+    else:
+        return np.mean(pred)
+
+# ==========================================
+# 3. PARALLEL API EXECUTION
+# ==========================================
+def call_worker(client, tensor_filepath):
+    """Pings a single Hugging Face API Worker"""
+    try:
+        result = client.predict(tensor_file=handle_file(tensor_filepath), api_name="/predict")
+        # Result should be a dictionary: {"predictions": [...]}
+        preds = result.get("predictions", [])
+        if not preds:
+            return 0.5 # Default middle ground if error
+        return confident_strategy(preds)
+    except Exception as e:
+        print(f"API Call Failed: {e}")
+        return 0.5 
+
+def analyze_video(video_path):
+    if not video_path:
+        return "<div style='color:var(--red); font-family:Syne;'>Please upload a video file.</div>"
+    
+    # 1. Extract Faces locally
+    input_size = 380
+    faces = face_extractor.process_video(video_path, frames_per_video=16)
+    
+    if len(faces) == 0:
+        return "<div style='color:var(--amber); font-family:Syne; padding:20px;'>No faces detected. Please upload a clear video.</div>"
+
+    x = []
+    for frame_data in faces:
+        for face in frame_data["faces"]:
+            resized_face = isotropically_resize_image(face, input_size)
+            resized_face = put_to_center(resized_face, input_size)
+            x.append(resized_face)
+            if len(x) >= 16 * 4: 
+                break
+
+    x = np.array(x, dtype=np.uint8)
+    x = torch.tensor(x, device=device).float()
+    x = x.permute((0, 3, 1, 2))
+    for i in range(len(x)):
+        x[i] = normalize_transform(x[i] / 255.)
+
+    # 2. Save the math to a temporary file
+    temp_dir = tempfile.gettempdir()
+    tensor_path = os.path.join(temp_dir, "batch_tensor.pt")
+    torch.save(x, tensor_path)
+
+    # 3. Ping all 7 Workers in parallel
+    worker_scores = []
+    with concurrent.futures.ThreadPoolExecutor(max_workers=7) as executor:
+        futures = [executor.submit(call_worker, client, tensor_path) for client in clients]
+        for future in concurrent.futures.as_completed(futures):
+            worker_scores.append(future.result())
+
+    # 4. Aggregate results
+    final_score = np.mean(worker_scores)
+    is_fake = final_score > 0.5
+    display_score = (final_score * 100) if is_fake else ((1 - final_score) * 100)
+    
+    # Format the individual scores for the UI
+    model_bars_html = ""
+    for i, score in enumerate(worker_scores):
+        percentage = score * 100
+        color = "var(--red)" if percentage > 50 else "var(--green)"
+        model_bars_html += f"""
+        <div class="metric-row">
+            <div class="metric-header"><span class="metric-name">EfficientNet Node {i+1}</span><span class="metric-value">{percentage:.1f}%</span></div>
+            <div class="metric-bar"><div class="metric-fill" style="width:{percentage}%; background:{color}"></div></div>
+        </div>
+        """
+
+    # 5. Inject into your Custom HTML Template
+    verdict_color = "var(--red)" if is_fake else "var(--green)"
+    verdict_text = "DEEPFAKE DETECTED" if is_fake else "AUTHENTIC CONTENT"
+    verdict_desc = "High confidence manipulation detected. Neural forensics indicate spatial anomalies and blending artifacts typical of synthetic face-swapping." if is_fake else "No significant facial manipulation detected. Spatial forensics are within normal parameters. Content appears to be authentic media."
+    
+    # Calculate a proxy for "Face Anomaly" vs "Temporal" based on the raw score to fill your template's visual metrics
+    face_anomaly_score = (final_score * 100) if is_fake else (final_score * 100)
+    
+    html_report = f"""
+    <div class="report-layout">
+        <div class="report-card accent">
+            <div class="card-title"><span class="dot"></span>Forensic Analysis Report</div>
+            <div style="margin-top:8px">
+                <div style="display:flex;justify-content:space-between;align-items:center;margin-bottom:24px">
+                    <div>
+                        <div style="font-family:'JetBrains Mono',monospace;font-size:10px;letter-spacing:2px;color:var(--text-faint);text-transform:uppercase">Verdict</div>
+                        <div style="font-family:'Bebas Neue',sans-serif;font-size:32px;color:{verdict_color};margin-top:4px">{verdict_text}</div>
+                    </div>
+                    <div style="text-align:right">
+                        <div style="font-family:'Bebas Neue',sans-serif;font-size:48px;color:{verdict_color};text-shadow:0 0 20px {verdict_color};line-height:1">{display_score:.1f}%</div>
+                        <div style="font-family:'JetBrains Mono',monospace;font-size:9px;letter-spacing:2px;color:{verdict_color};text-transform:uppercase">Confidence</div>
+                    </div>
+                </div>
+                <p style="color:var(--text-dim); font-size:14px; line-height:1.6; margin-bottom:20px;">{verdict_desc}</p>
+                <ul class="forensic-list">
+                    <li class="forensic-item">
+                        <div class="forensic-icon"><svg viewBox="0 0 24 24"><circle cx="12" cy="8" r="4"/><path d="M20 21a8 8 0 1 0-16 0"/></svg></div>
+                        <span class="forensic-name">Spatial Artifact Detection</span>
+                        <span class="forensic-status {'alert' if is_fake else 'pass'}">{'Anomaly' if is_fake else 'Pass'}</span>
+                    </li>
+                    <li class="forensic-item">
+                        <div class="forensic-icon"><svg viewBox="0 0 24 24"><path d="M4 15s1-1 4-1 5 2 8 2 4-1 4-1V3s-1 1-4 1-5-2-8-2-4 1-4 1z"/><line x1="4" y1="22" x2="4" y2="15"/></svg></div>
+                        <span class="forensic-name">Feature Extraction Integrity</span>
+                        <span class="forensic-status {'alert' if face_anomaly_score > 60 else 'pass'}">{'Fail' if face_anomaly_score > 60 else 'Normal'}</span>
+                    </li>
+                </ul>
+            </div>
+        </div>
+
+        <div style="display:flex;flex-direction:column;gap:2px">
+            <div class="report-card" style="flex:1">
+                <div class="card-title"><span class="dot"></span>Ensemble Node Breakdown</div>
+                <div style="margin-top:16px">
+                    {model_bars_html}
+                </div>
+            </div>
+        </div>
+    </div>
+    """
+    return html_report
+
+# ==========================================
+# 4. MASTER UI - NETFLIX HTML INTEGRATION
+# ==========================================
+# We pull your exact CSS variables and styling directly from your deepfake-detector.html
+css = """
+@import url('https://fonts.googleapis.com/css2?family=Bebas+Neue&family=Syne:wght@400;600;700;800&family=JetBrains+Mono:wght@300;400;500&display=swap');
+
+:root {
+    --bg: #030508;
+    --bg2: #070c12;
+    --panel: rgba(8, 18, 30, 0.85);
+    --border: rgba(0, 210, 255, 0.12);
+    --border-bright: rgba(0, 210, 255, 0.45);
+    --cyan: #00d2ff;
+    --red: #ff2d55;
+    --green: #00ff88;
+    --amber: #ffb800;
+    --text: #e8f4ff;
+    --text-dim: rgba(232, 244, 255, 0.5);
+    --text-faint: rgba(232, 244, 255, 0.25);
+}
+
+body, .gradio-container { background-color: var(--bg) !important; color: var(--text) !important; font-family: 'Syne', sans-serif !important; }
+.gr-panel { background: var(--panel) !important; border: 1px solid var(--border) !important; border-radius: 4px !important; }
+
+/* Dashboard Titles */
+.veridex-title { font-family: 'Bebas Neue', sans-serif; font-size: 60px; letter-spacing: 4px; color: var(--text); text-align: center; margin-top: 40px;}
+.veridex-title span { color: var(--cyan); }
+.veridex-sub { font-family: 'JetBrains Mono', monospace; font-size: 12px; letter-spacing: 2px; text-transform: uppercase; color: var(--cyan); text-align: center; margin-bottom: 40px; }
+
+/* Custom HTML injected classes from your design */
+.report-layout { display: grid; grid-template-columns: 1fr 1fr; gap: 16px; margin-top: 20px; }
+.report-card { background: var(--panel); border: 1px solid var(--border); padding: 30px; }
+.report-card.accent { border-color: rgba(0,210,255,0.2); background: rgba(0, 210, 255, 0.04); }
+.card-title { font-family: 'JetBrains Mono', monospace; font-size: 10px; letter-spacing: 3px; text-transform: uppercase; color: var(--cyan); margin-bottom: 16px; display: flex; align-items: center; gap: 8px; }
+.card-title .dot { width: 5px; height: 5px; border-radius: 50%; background: var(--cyan); box-shadow: 0 0 8px var(--cyan); }
+
+.forensic-list { list-style: none; display: flex; flex-direction: column; gap: 12px; padding:0; }
+.forensic-item { display: flex; align-items: center; gap: 12px; padding: 14px 16px; border: 1px solid var(--border); }
+.forensic-icon { width: 32px; height: 32px; border: 1px solid var(--border-bright); display: flex; align-items: center; justify-content: center; }
+.forensic-icon svg { width: 14px; height: 14px; stroke: var(--cyan); fill: none; stroke-width: 2; }
+.forensic-name { font-size: 13px; font-weight: 600; flex: 1; font-family: 'Syne', sans-serif;}
+.forensic-status { font-family: 'JetBrains Mono', monospace; font-size: 9px; letter-spacing: 2px; text-transform: uppercase; padding: 3px 8px; }
+.forensic-status.pass { color: var(--green); border: 1px solid rgba(0,255,136,0.3); background: rgba(0,255,136,0.05); }
+.forensic-status.alert { color: var(--red); border: 1px solid rgba(255,45,85,0.3); background: rgba(255,45,85,0.05); }
+
+.metric-row { margin-bottom: 14px; }
+.metric-header { display: flex; justify-content: space-between; margin-bottom: 6px; }
+.metric-name { font-family: 'JetBrains Mono', monospace; font-size: 10px; letter-spacing: 1.5px; text-transform: uppercase; color: var(--text-dim); }
+.metric-value { font-family: 'JetBrains Mono', monospace; font-size: 10px; color: var(--text); }
+.metric-bar { height: 3px; background: rgba(255,255,255,0.06); width: 100%; overflow: hidden; }
+.metric-fill { height: 100%; transition: width 1s ease; }
+
+@media (max-width: 900px) { .report-layout { grid-template-columns: 1fr; } }
+"""
+
+with gr.Blocks(css=css, theme=gr.themes.Default(neutral_hue="slate", primary_hue="cyan")) as app:
+    gr.HTML("""
+    <div class="veridex-title">VERI<span>DEX</span></div>
+    <div class="veridex-sub">Neural Detection Engine v4.2 // Distributed Architecture</div>
+    """)
+    
+    with gr.Row():
+        with gr.Column(scale=1):
+            gr.Markdown("### 1. Ingest Video Evidence")
+            video_in = gr.Video(label="Upload Media (.mp4, .avi)")
+            analyze_btn = gr.Button("Run Distributed Ensemble Analysis", variant="primary", size="lg")
+            
+            gr.HTML("""
+            <div style="margin-top:20px; font-family:'JetBrains Mono'; font-size:10px; color:var(--text-faint); line-height:1.8;">
+                › Local MTCNN Node Active<br>
+                › 7 Parallel EfficientNet Endpoints Linked<br>
+                › Awaiting input...
+            </div>
+            """)
+            
+        with gr.Column(scale=2):
+            gr.Markdown("### 2. Forensic Output")
+            report_out = gr.HTML(value="<div style='color:var(--text-dim); padding:40px; text-align:center; border:1px dashed var(--border);'>Awaiting video analysis...</div>")
+            
+    analyze_btn.click(fn=analyze_video, inputs=video_in, outputs=report_out)
+
+if __name__ == "__main__":
+    app.launch()