APEXPRIT commited on
Commit
d7c4808
·
verified ·
1 Parent(s): cb0dba7

Upload folder using huggingface_hub

Browse files
Files changed (1) hide show
  1. api.py +61 -0
api.py CHANGED
@@ -43,6 +43,12 @@ async def startup_event():
43
  if os.path.exists(model_path):
44
  pipeline.load_model(model_path)
45
  logger.info("Pre-trained model loaded successfully.")
 
 
 
 
 
 
46
  else:
47
  logger.warning(f"No pre-trained model found at {model_path}. You must generate it first!")
48
 
@@ -50,6 +56,61 @@ async def startup_event():
50
  except Exception as e:
51
  logger.error(f"Failed to initialize pipeline: {e}")
52
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
53
  @app.post("/analyze")
54
  async def analyze_video_endpoint(
55
  file: UploadFile = File(...),
 
43
  if os.path.exists(model_path):
44
  pipeline.load_model(model_path)
45
  logger.info("Pre-trained model loaded successfully.")
46
+
47
+ # Also load calibrator profiles
48
+ calib_path = os.path.join("models", "subject_calibrator.json")
49
+ if os.path.exists(calib_path):
50
+ pipeline.calibrator.load_profiles(calib_path)
51
+ logger.info(f"Loaded calibration profiles from {calib_path}")
52
  else:
53
  logger.warning(f"No pre-trained model found at {model_path}. You must generate it first!")
54
 
 
56
  except Exception as e:
57
  logger.error(f"Failed to initialize pipeline: {e}")
58
 
59
+ @app.post("/calibrate")
60
+ async def calibrate_subject_endpoint(
61
+ file: UploadFile = File(...),
62
+ subject_id: str = Form(...)
63
+ ):
64
+ if not pipeline:
65
+ raise HTTPException(status_code=500, detail="ML Pipeline not initialized yet")
66
+
67
+ # Create temporary directory
68
+ temp_dir = tempfile.mkdtemp(prefix="deception_calib_")
69
+ video_path = os.path.join(temp_dir, file.filename)
70
+
71
+ try:
72
+ # Save video
73
+ with open(video_path, "wb") as f:
74
+ content = await file.read()
75
+ f.write(content)
76
+
77
+ logger.info(f"CALIBRATING subject: {subject_id}")
78
+
79
+ # Extract features from the TRUTHFUL video
80
+ facial_df = pipeline.facial_extractor.process_video(video_path)
81
+
82
+ # Extract audio features
83
+ audio_temp = os.path.join(temp_dir, "audio.wav")
84
+ from src.utils.io import extract_audio_from_video
85
+ audio_path = extract_audio_from_video(video_path, audio_temp)
86
+ audio_df = pipeline.audio_extractor.process_audio_segments(str(audio_path), segment_duration=2.0)
87
+
88
+ # Fuse
89
+ aligned_df = pipeline.fusion.align_features(facial_df, audio_df, video_fps=pipeline.config.features.fps)
90
+ fused_df = pipeline.fusion.extract_fusion_features(aligned_df)
91
+
92
+ # Add to subject calibrator
93
+ pipeline.calibrator.add_subject_profile(subject_id, fused_df)
94
+
95
+ # Save profiles to disk so they persist after restart
96
+ calib_path = os.path.join("models", "subject_calibrator.json")
97
+ pipeline.calibrator.save_profiles(calib_path)
98
+
99
+ return {
100
+ "status": "success",
101
+ "message": f"Calibration complete for {subject_id}",
102
+ "frames_analyzed": len(fused_df)
103
+ }
104
+
105
+ except Exception as e:
106
+ logger.error(f"Calibration failed: {str(e)}")
107
+ raise HTTPException(status_code=500, detail=f"Calibration failed: {str(e)}")
108
+ finally:
109
+ try:
110
+ if os.path.exists(video_path): os.remove(video_path)
111
+ os.rmdir(temp_dir)
112
+ except: pass
113
+
114
  @app.post("/analyze")
115
  async def analyze_video_endpoint(
116
  file: UploadFile = File(...),