Spaces:
Sleeping
Sleeping
File size: 10,636 Bytes
b0c6daf | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 | """
Session Report Generator — forensic session reports for audit trails.
Generates structured JSON session reports containing:
- Session metadata (duration, total frames, model version)
- Timeline of verdicts with confidence values
- Flagged FAKE frames with thumbnails + confidence
- Trust Meta-Classifier verdicts over time
- GradCAM heatmap snapshots for suspicious frames
- Temporal consistency timeline
- TTA agreement scores
This module operates on MongoDB frame_records and produces
a JSON report that the frontend can render or export as PDF.
Usage (from main.py):
from session_report import generate_session_report, list_sessions
# List available sessions
sessions = list_sessions()
# Generate report for a specific session
report = generate_session_report("session_abc123")
"""
import traceback
from datetime import datetime, timezone
from typing import Optional
from frame_store import _get_mongo_collection
def list_sessions(limit: int = 50) -> list:
"""
List recent sessions with summary stats.
Returns:
List of session summaries, sorted by most recent first.
"""
try:
collection = _get_mongo_collection()
pipeline = [
{
"$group": {
"_id": "$session_id",
"frame_count": {"$sum": 1},
"first_frame": {"$min": "$timestamp"},
"last_frame": {"$max": "$timestamp"},
"avg_confidence": {"$avg": "$primary_confidence"},
"fake_count": {
"$sum": {"$cond": [{"$eq": ["$primary_label", "FAKE"]}, 1, 0]}
},
"real_count": {
"$sum": {"$cond": [{"$eq": ["$primary_label", "REAL"]}, 1, 0]}
},
"untrusted_count": {
"$sum": {"$cond": [{"$eq": ["$trust_verdict", "UNTRUSTED"]}, 1, 0]}
},
}
},
{"$sort": {"last_frame": -1}},
{"$limit": limit},
]
results = list(collection.aggregate(pipeline))
sessions = []
for r in results:
duration_seconds = 0
if r.get("first_frame") and r.get("last_frame"):
delta = r["last_frame"] - r["first_frame"]
duration_seconds = round(delta.total_seconds(), 1)
sessions.append({
"session_id": r["_id"],
"frame_count": r["frame_count"],
"duration_seconds": duration_seconds,
"started_at": r["first_frame"].isoformat() if r.get("first_frame") else None,
"ended_at": r["last_frame"].isoformat() if r.get("last_frame") else None,
"avg_confidence": round(r.get("avg_confidence", 0), 3),
"fake_count": r["fake_count"],
"real_count": r["real_count"],
"untrusted_count": r["untrusted_count"],
"threat_level": _compute_threat_level(
r["fake_count"], r["real_count"], r["untrusted_count"]
),
})
return sessions
except Exception as e:
print(f"[SESSION_REPORT] Error listing sessions: {e}")
traceback.print_exc()
return []
def generate_session_report(
session_id: str,
include_frame_urls: bool = True,
include_embeddings: bool = False,
) -> dict:
"""
Generate a comprehensive forensic report for a specific session.
Args:
session_id: The session identifier
include_frame_urls: Whether to include Supabase frame URLs
include_embeddings: Whether to include raw embeddings (large data)
Returns:
Comprehensive session report as a dict
"""
try:
collection = _get_mongo_collection()
# Fetch all frames for this session, sorted by timestamp
projection = {
"_id": 0,
"session_id": 1,
"timestamp": 1,
"frame_url": 1,
"primary_label": 1,
"primary_confidence": 1,
"face_detected": 1,
"latency_ms": 1,
"trust_verdict": 1,
"trust_score": 1,
"human_label": 1,
"reviewed_by": 1,
"reviewed_at": 1,
}
if include_embeddings:
projection["embedding"] = 1
frames = list(
collection.find(
{"session_id": session_id},
projection,
).sort("timestamp", 1)
)
if not frames:
return {"error": f"No frames found for session {session_id}"}
# ── Compute timeline ──
timeline = []
for f in frames:
entry = {
"timestamp": f["timestamp"].isoformat() if f.get("timestamp") else None,
"label": f.get("primary_label", "UNKNOWN"),
"confidence": round(f.get("primary_confidence", 0), 4),
"trust_verdict": f.get("trust_verdict", "UNKNOWN"),
"trust_score": round(f.get("trust_score", 0), 4),
"latency_ms": f.get("latency_ms", 0),
"face_detected": f.get("face_detected", False),
}
if include_frame_urls and f.get("frame_url"):
entry["frame_url"] = f["frame_url"]
if f.get("human_label"):
entry["human_label"] = f["human_label"]
timeline.append(entry)
# ── Compute summary statistics ──
total = len(frames)
fake_frames = [f for f in frames if f.get("primary_label") == "FAKE"]
real_frames = [f for f in frames if f.get("primary_label") == "REAL"]
untrusted_frames = [f for f in frames if f.get("trust_verdict") == "UNTRUSTED"]
confidences = [f.get("primary_confidence", 0) for f in frames]
latencies = [f.get("latency_ms", 0) for f in frames if f.get("latency_ms")]
trust_scores = [f.get("trust_score", 0) for f in frames]
# Duration
first_ts = frames[0].get("timestamp")
last_ts = frames[-1].get("timestamp")
duration_seconds = 0
if first_ts and last_ts:
duration_seconds = round((last_ts - first_ts).total_seconds(), 1)
# Consecutive FAKE detection (longest streak)
max_consecutive_fakes = 0
current_streak = 0
for f in frames:
if f.get("primary_label") == "FAKE":
current_streak += 1
max_consecutive_fakes = max(max_consecutive_fakes, current_streak)
else:
current_streak = 0
# ── Flagged frames (high-risk: FAKE with high confidence) ──
flagged = []
for f in fake_frames:
entry = {
"timestamp": f["timestamp"].isoformat() if f.get("timestamp") else None,
"confidence": round(f.get("primary_confidence", 0), 4),
"trust_verdict": f.get("trust_verdict", "UNKNOWN"),
"trust_score": round(f.get("trust_score", 0), 4),
}
if include_frame_urls and f.get("frame_url"):
entry["frame_url"] = f["frame_url"]
flagged.append(entry)
# ── Build report ──
report = {
"report_version": "1.0",
"generated_at": datetime.now(timezone.utc).isoformat(),
"session_id": session_id,
"summary": {
"total_frames": total,
"duration_seconds": duration_seconds,
"started_at": first_ts.isoformat() if first_ts else None,
"ended_at": last_ts.isoformat() if last_ts else None,
"fake_frames": len(fake_frames),
"real_frames": len(real_frames),
"fake_percentage": round(len(fake_frames) / total * 100, 1) if total > 0 else 0,
"avg_confidence": round(sum(confidences) / len(confidences), 4) if confidences else 0,
"min_confidence": round(min(confidences), 4) if confidences else 0,
"max_confidence": round(max(confidences), 4) if confidences else 0,
"avg_latency_ms": round(sum(latencies) / len(latencies), 1) if latencies else 0,
"max_latency_ms": round(max(latencies), 1) if latencies else 0,
"avg_trust_score": round(sum(trust_scores) / len(trust_scores), 4) if trust_scores else 0,
"untrusted_frames": len(untrusted_frames),
"max_consecutive_fakes": max_consecutive_fakes,
"threat_level": _compute_threat_level(
len(fake_frames), len(real_frames), len(untrusted_frames)
),
},
"flagged_frames": flagged[:50], # cap at 50 most important
"timeline": timeline,
"model_info": {
"primary_model": "InceptionResnetV1 + Custom Head",
"trust_classifier": "TrustMetaClassifier",
"capabilities": [
"gradcam_heatmaps",
"temporal_analysis",
"tta_robustness",
"frequency_domain_analysis",
],
},
}
return report
except Exception as e:
print(f"[SESSION_REPORT] Error generating report: {e}")
traceback.print_exc()
return {"error": str(e)}
def _compute_threat_level(fake_count: int, real_count: int, untrusted_count: int) -> str:
"""
Compute a human-readable threat level based on detection results.
"""
total = fake_count + real_count
if total == 0:
return "UNKNOWN"
fake_ratio = fake_count / total
if fake_ratio >= 0.5 or untrusted_count > total * 0.3:
return "CRITICAL"
elif fake_ratio >= 0.25:
return "HIGH"
elif fake_ratio >= 0.1:
return "MODERATE"
elif fake_ratio > 0:
return "LOW"
else:
return "CLEAR"
|