deepfake-backend / server.py
Devanshu2025's picture
HF files and audio detection
bdc74df
from fastapi import FastAPI, File, UploadFile, Form, HTTPException
from fastapi.middleware.cors import CORSMiddleware
from fastapi.responses import JSONResponse
import os
import tempfile
import shutil
from dotenv import load_dotenv
from model_utils import load_model, get_device
from preprocessing import preprocess_video, predict
# Audio deepfake detection imports (separate from video pipeline)
from audio_predict import predict_audio, AudioPredictionError
from audio_preprocessing import AudioValidationError, AudioLoadError
# Load environment variables
load_dotenv()
# Initialize FastAPI app
app = FastAPI(
title="Deepfake Detection API",
description="Video and Audio deepfake detection API",
version="1.0.0"
)
# CORS configuration for production-ready deployment
# Get frontend URL from environment variable (blank for development)
frontend_url = os.getenv("FRONTEND_URL", "").strip()
# Build CORS allowed origins list
allowed_origins = [
"http://localhost:8080", # Vite dev server default
"http://localhost:5173", # Alternative Vite port
"http://127.0.0.1:8080",
"http://127.0.0.1:5173",
]
# Add production frontend URL if specified
if frontend_url:
allowed_origins.append(frontend_url)
print(f"✓ Production frontend URL added to CORS: {frontend_url}")
else:
print("✓ Development mode: Using localhost CORS origins")
# Configure CORS middleware
app.add_middleware(
CORSMiddleware,
allow_origins=allowed_origins,
allow_credentials=True,
allow_methods=["*"],
allow_headers=["*"],
)
print("\n=== Deepfake Detection Server Configuration ===")
print(f"Allowed CORS origins: {allowed_origins}")
print(f"Device: {get_device()}")
print("=" * 50 + "\n")
@app.get("/")
async def root():
"""Health check endpoint"""
return {
"status": "online",
"service": "Deepfake Detection API",
"version": "1.1.0",
"device": get_device(),
"capabilities": ["video", "audio"]
}
@app.post("/api/predict")
async def predict_video_endpoint(
file: UploadFile = File(...),
sequence_length: int = Form(...),
face_focus: bool = Form(True)
):
"""
Predict whether a video is real or fake.
Args:
file: Video file to analyze
sequence_length: Number of frames to extract (10, 20, 40, 60, 80, 100)
face_focus: Whether to focus on faces (currently always enabled)
Returns:
JSON response with prediction result and confidence
"""
temp_video_path = None
try:
# Validate sequence length
valid_lengths = [10, 20, 40, 60, 80, 100]
if sequence_length not in valid_lengths:
raise HTTPException(
status_code=400,
detail=f"Invalid sequence_length. Must be one of {valid_lengths}"
)
# Validate file type
if not file.content_type or not file.content_type.startswith('video/'):
raise HTTPException(
status_code=400,
detail="File must be a video"
)
print(f"\n{'='*50}")
print(f"Processing video: {file.filename}")
print(f"Sequence length: {sequence_length} frames")
print(f"Face focus: {face_focus}")
print(f"{'='*50}\n")
# Save uploaded video to temporary file
with tempfile.NamedTemporaryFile(delete=False, suffix=os.path.splitext(file.filename)[1]) as temp_file:
shutil.copyfileobj(file.file, temp_file)
temp_video_path = temp_file.name
print(f"✓ Video saved to: {temp_video_path}")
# Load model for the specified sequence length
device = get_device()
model = load_model(sequence_length, device)
if model is None:
raise HTTPException(
status_code=500,
detail=f"Failed to load model for {sequence_length} frames"
)
print(f"✓ Model loaded successfully")
# Preprocess video
print(f"⏳ Preprocessing video...")
frames_tensor, preprocessed_images, face_cropped_images, faces_found = preprocess_video(
temp_video_path,
sequence_length,
save_preprocessed=False # Set to True if you want to save frames
)
print(f"✓ Preprocessing complete")
# Make prediction
print(f"⏳ Running prediction...")
prediction_int, confidence = predict(model, frames_tensor, device)
# Convert prediction to label
prediction_label = "REAL" if prediction_int == 1 else "FAKE"
print(f"\n{'='*50}")
print(f"✓ PREDICTION: {prediction_label}")
print(f"✓ CONFIDENCE: {confidence:.1f}%")
print(f"{'='*50}\n")
# Return response with frame images for frontend display
# Limit to max 6 frames to keep response size reasonable
display_frames = face_cropped_images[:6] if len(face_cropped_images) > 6 else face_cropped_images
return JSONResponse(content={
"prediction": prediction_label,
"confidence": round(confidence, 1),
"sequence_length": sequence_length,
"device": device,
"faces_found": faces_found,
"total_frames_analyzed": len(face_cropped_images),
"frame_images": display_frames
})
except HTTPException:
raise
except Exception as e:
print(f"\n❌ Error during prediction: {str(e)}")
import traceback
traceback.print_exc()
raise HTTPException(
status_code=500,
detail=f"Prediction failed: {str(e)}"
)
finally:
# Clean up temporary file
if temp_video_path and os.path.exists(temp_video_path):
try:
os.unlink(temp_video_path)
print(f"✓ Cleaned up temporary file")
except Exception as e:
print(f"⚠ Warning: Could not delete temporary file: {e}")
# =============================================================================
# AUDIO DEEPFAKE DETECTION ENDPOINT
# =============================================================================
@app.post("/api/audio/predict")
async def predict_audio_endpoint(file: UploadFile = File(...)):
"""
Predict whether an audio file is real or fake (deepfake).
Uses MelodyMachine/Deepfake-audio-detection-V2 (Wav2Vec2-based model).
Args:
file: Audio file to analyze (WAV, MP3, FLAC, M4A, OGG supported)
Returns:
JSON response with prediction result and confidence:
{
"prediction": "REAL" | "FAKE",
"confidence": float (0-100),
"model": "MelodyMachine/Deepfake-audio-detection-V2",
"all_scores": {"real": float, "fake": float}
}
"""
temp_audio_path = None
try:
# Validate content type
if file.content_type and not file.content_type.startswith('audio/'):
raise HTTPException(
status_code=400,
detail="File must be an audio file"
)
print(f"\n{'='*50}")
print(f"Processing audio: {file.filename}")
print(f"Content type: {file.content_type}")
print(f"{'='*50}\n")
# Get file extension from filename
file_ext = os.path.splitext(file.filename)[1] if file.filename else '.wav'
# Save uploaded audio to temporary file
with tempfile.NamedTemporaryFile(delete=False, suffix=file_ext) as temp_file:
shutil.copyfileobj(file.file, temp_file)
temp_audio_path = temp_file.name
print(f"✓ Audio saved to: {temp_audio_path}")
# Run audio prediction
print(f"⏳ Running audio deepfake detection...")
result = predict_audio(temp_audio_path, file.content_type)
print(f"\n{'='*50}")
print(f"✓ AUDIO PREDICTION: {result['prediction']}")
print(f"✓ CONFIDENCE: {result['confidence']:.1f}%")
print(f"{'='*50}\n")
return JSONResponse(content=result)
except AudioValidationError as e:
print(f"\n❌ Audio validation error: {str(e)}")
raise HTTPException(status_code=400, detail=str(e))
except AudioLoadError as e:
print(f"\n❌ Audio load error: {str(e)}")
raise HTTPException(status_code=400, detail=str(e))
except AudioPredictionError as e:
print(f"\n❌ Audio prediction error: {str(e)}")
raise HTTPException(status_code=500, detail=str(e))
except HTTPException:
raise
except Exception as e:
print(f"\n❌ Error during audio prediction: {str(e)}")
import traceback
traceback.print_exc()
raise HTTPException(
status_code=500,
detail=f"Audio prediction failed: {str(e)}"
)
finally:
# Clean up temporary file
if temp_audio_path and os.path.exists(temp_audio_path):
try:
os.unlink(temp_audio_path)
print(f"✓ Cleaned up temporary audio file")
except Exception as e:
print(f"⚠ Warning: Could not delete temporary audio file: {e}")
@app.get("/api/models")
async def list_available_models():
"""List all available models and their frame counts"""
import glob
models_dir = "models"
model_files = glob.glob(os.path.join(models_dir, "*.pt"))
models_info = []
for model_path in model_files:
filename = os.path.basename(model_path)
try:
parts = filename.split("_")
accuracy = parts[1]
frames = parts[3]
models_info.append({
"filename": filename,
"frames": int(frames),
"accuracy": f"{accuracy}%"
})
except (IndexError, ValueError):
continue
return {
"available_models": sorted(models_info, key=lambda x: x["frames"]),
"total": len(models_info)
}
if __name__ == "__main__":
import uvicorn
port = int(os.getenv("PORT", 8000))
uvicorn.run(app, host="0.0.0.0", port=port)