Update app.py

app.py

@@ -1,19 +1,20 @@
-import gradio as gr
+import os
 import torch
 import torch.nn as nn
 import numpy as np
 from torchvision import transforms as T
 from torchvision.transforms.v2 import ToDtype
 from decord import VideoReader, cpu
-from trainers import vificlip
-from utils.config import get_config
-from utils.logger import create_logger
+import gradio as gr
 
 # -------------------------
-# Setup Device & Seed
+# Step 0: Download model from Google Drive if not exists
 # -------------------------
-device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
-torch.manual_seed(42)
+model_path = 'vifi_clip_30_epochs_k400_full_finetuned.pth'
+if not os.path.exists(model_path):
+    print(f"🔽 Downloading model to {model_path}...")
+    os.system("pip install -q gdown")
+    os.system("gdown --id 1Nx30Kbu5xnv6dPwz4I3Ivy380LCdp1Md -O vifi_clip_30_epochs_k400_full_finetuned.pth")
 
 # -------------------------
 # Transform
@@ -34,16 +35,22 @@ class ClassificationHead(nn.Module):
     def __init__(self, input_dim=512, num_classes=1):
         super().__init__()
         self.dense = nn.Linear(input_dim, num_classes)
+
     def forward(self, x):
         return self.dense(x)
 
 # -------------------------
-# Load ViFi-CLIP + Classifier
+# Load ViFi-CLIP Model
 # -------------------------
+from trainers import vificlip
+from utils.config import get_config
+from utils.logger import create_logger
+
 cfgpth = 'configs/zero_shot/train/k400/16_16_vifi_clip.yaml'
-model_path = 'vifi_clip_30_epochs_k400_full_finetuned.pth'
 classifier_path = 'best_detector_model.pt'
 
+device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+
 class parse_option:
     def __init__(self):
         self.config = cfgpth
@@ -69,7 +76,7 @@ classifier.to(device)
 classifier.eval()
 
 # -------------------------
-# Inference Function (with threshold)
+# Inference Function
 # -------------------------
 def predict_video(video_path, threshold=0.5):
     preprocess = _transform(224)
@@ -104,15 +111,15 @@ def predict_video(video_path, threshold=0.5):
         return f"❌ Error: {str(e)}"
 
 # -------------------------
-# Gradio UI (with slider)
+# Gradio UI
 # -------------------------
 gr.Interface(
     fn=predict_video,
     inputs=[
-        gr.Video(type="filepath", label="Upload Video"),
+        gr.Video(type="filepath", label="Upload Video (.mp4)"),
         gr.Slider(0.0, 1.0, value=0.5, step=0.01, label="Threshold (Real ≥ Threshold)")
     ],
     outputs="text",
-    title="Fake Video Detection with Threshold Control",
-    description="Upload a video to classify it as Real or Fake. Adjust the threshold to tune sensitivity."
+    title="🧠 Deepfake Detection with ViFi-CLIP",
+    description="Upload a video to classify it as Real or Fake. Threshold slider lets you adjust sensitivity."
 ).launch()