Spaces:
Sleeping
Sleeping
File size: 6,288 Bytes
6903fe1 | 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 | import asyncio
from pathlib import Path
from fastapi import FastAPI, File, Form, UploadFile
from src.video.video_processor import VideoAnalyzer
app = FastAPI(title="Video Prompt Detection API")
_ANALYZER: VideoAnalyzer | None = None
@app.on_event("startup")
def load_analyzer() -> None:
global _ANALYZER
if _ANALYZER is None:
_ANALYZER = VideoAnalyzer()
@app.get("/")
def health_check() -> dict:
return {"status": "ok", "engine": "video"}
@app.post("/analyze_video")
async def analyze_video(
video: UploadFile = File(...),
audio_transcript: str = Form(""),
target_fps: float = Form(5.0),
max_frames: int | None = Form(None),
run_injection: bool = Form(True),
run_cross_modal: bool = Form(True),
run_caption: bool = Form(True),
run_vision_deepfake: bool = Form(True),
run_avsync: bool = Form(True),
log_frames: bool = Form(True),
) -> dict:
if _ANALYZER is None:
load_analyzer()
analyzer = _ANALYZER
video_bytes = await video.read()
log_path = None
if log_frames:
log_path = f"/tmp/video_frame_log_{int(asyncio.get_event_loop().time()*1000)}.jsonl"
frames, summary = analyzer.analyze_video_bytes(
video_bytes,
audio_transcript=audio_transcript,
target_fps=target_fps,
max_frames=max_frames,
run_injection=run_injection,
run_cross_modal=run_cross_modal,
run_caption=run_caption,
run_vision_deepfake=run_vision_deepfake,
run_avsync=run_avsync,
log_path=Path(log_path) if log_path else None,
)
top_risky = sorted(frames, key=lambda f: f.final_score, reverse=True)[:5]
def _action_from_score(score: float) -> str:
if score >= 0.7:
return "BLOCK"
if score >= 0.5:
return "FLAG"
return "ALLOW"
def flatten(frame):
action = _action_from_score(frame.final_score)
return {
"frame_index": frame.frame_index,
"timestamp_sec": frame.timestamp_sec,
"final_score": frame.final_score,
"action": action,
"deepfake_score": frame.deepfake_score,
"deepfake_label": frame.deepfake_label,
"deepfake_is_fake": frame.deepfake_is_fake,
"injection_risk": frame.injection.get("risk_score", 0.0),
"injection_reason": frame.injection.get("reason", ""),
"cross_modal_score": frame.cross_modal.get("consistency_score", 0.0),
"ocr_vs_image_score": frame.ocr_vs_image.get("consistency_score", 0.0),
"caption_alignment_score": frame.caption_alignment.get("alignment_score", 0.0),
"caption": frame.caption_alignment.get("caption", ""),
"ocr_text": frame.ocr_text,
}
action = _action_from_score(summary.get("max_final_score", 0.0))
explanations = [
f"avg_deepfake={summary.get('avg_deepfake_score', 0.0)}",
f"avsync={summary.get('avsync_score', 0.0)}",
f"max_final={summary.get('max_final_score', 0.0)}",
]
return {
"summary": summary,
"timeline": [f.__dict__ for f in frames],
"timeline_flat": [flatten(f) for f in frames],
"top_risky_frames": [f.__dict__ for f in top_risky],
"top_risky_frames_flat": [flatten(f) for f in top_risky],
"action": action,
"explanations": explanations,
"log_path": log_path,
}
@app.post("/analyze_webcam")
async def analyze_webcam(
camera_index: int = Form(0),
duration_sec: float = Form(10.0),
target_fps: float = Form(5.0),
run_injection: bool = Form(True),
run_cross_modal: bool = Form(True),
run_caption: bool = Form(True),
run_vision_deepfake: bool = Form(True),
run_avsync: bool = Form(True),
log_frames: bool = Form(True),
) -> dict:
if _ANALYZER is None:
load_analyzer()
analyzer = _ANALYZER
log_path = None
if log_frames:
log_path = f"/tmp/webcam_frame_log_{int(asyncio.get_event_loop().time()*1000)}.jsonl"
frames, summary = analyzer.analyze_webcam(
camera_index=camera_index,
duration_sec=duration_sec,
target_fps=target_fps,
run_injection=run_injection,
run_cross_modal=run_cross_modal,
run_caption=run_caption,
run_vision_deepfake=run_vision_deepfake,
run_avsync=run_avsync,
log_path=Path(log_path) if log_path else None,
)
top_risky = sorted(frames, key=lambda f: f.final_score, reverse=True)[:5]
def _action_from_score(score: float) -> str:
if score >= 0.7:
return "BLOCK"
if score >= 0.5:
return "FLAG"
return "ALLOW"
def flatten(frame):
action = _action_from_score(frame.final_score)
return {
"frame_index": frame.frame_index,
"timestamp_sec": frame.timestamp_sec,
"final_score": frame.final_score,
"action": action,
"deepfake_score": frame.deepfake_score,
"deepfake_label": frame.deepfake_label,
"deepfake_is_fake": frame.deepfake_is_fake,
"injection_risk": frame.injection.get("risk_score", 0.0),
"injection_reason": frame.injection.get("reason", ""),
"cross_modal_score": frame.cross_modal.get("consistency_score", 0.0),
"ocr_vs_image_score": frame.ocr_vs_image.get("consistency_score", 0.0),
"caption_alignment_score": frame.caption_alignment.get("alignment_score", 0.0),
"caption": frame.caption_alignment.get("caption", ""),
"ocr_text": frame.ocr_text,
}
action = _action_from_score(summary.get("max_final_score", 0.0))
explanations = [
f"avg_deepfake={summary.get('avg_deepfake_score', 0.0)}",
f"avsync={summary.get('avsync_score', 0.0)}",
f"max_final={summary.get('max_final_score', 0.0)}",
]
return {
"summary": summary,
"timeline": [f.__dict__ for f in frames],
"timeline_flat": [flatten(f) for f in frames],
"top_risky_frames": [f.__dict__ for f in top_risky],
"top_risky_frames_flat": [flatten(f) for f in top_risky],
"action": action,
"explanations": explanations,
"log_path": log_path,
}
|