Initial public release of V58 face anonymization evaluation

.gitattributes

@@ -58,3 +58,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 # Video files - compressed
 *.mp4 filter=lfs diff=lfs merge=lfs -text
 *.webm filter=lfs diff=lfs merge=lfs -text
+technical_report.pdf filter=lfs diff=lfs merge=lfs -text

MODEL_CARD.md

@@ -0,0 +1,57 @@
+# Model Card
+
+**Origin Data Lab** — privacy processing and Human GT delivery for real-world video data.
+
+## Evaluation context
+
+- Engine version: `V55-P0-FACE-ONLY`
+- Pipeline version: `v55-p0-face-only`
+- Location: Not specified
+- Environment: Not specified
+
+## Processing metrics
+
+| Metric | Value |
+|---|---:|
+| Resolution | 1280 × 720 |
+| Processed frames | 2,661 |
+| Source FPS | 30.000 |
+| Processing FPS | 11.305 |
+| Elapsed seconds | 235.393 |
+| Detected faces | 10,634 |
+| Blurred faces | 10,634 |
+| Rejected regions | 0 |
+| Mean confidence | 0.6373 |
+
+### Processing speed
+
+![Processing speed](charts/fps_chart.png)
+
+### Detection activity
+
+![Detection activity](charts/detection_chart.png)
+
+### Confidence distribution
+
+![Confidence distribution](charts/confidence_chart.png)
+
+## Review and intended use
+
+Human QA is recommended before production delivery.
+
+This report documents an internal engineering evaluation. It is not an independently audited accuracy benchmark.
+
+## Known limitations
+
+- Very small or distant faces
+- Severe occlusion
+- Motion blur and compression artifacts
+- Difficult low-light scenes
+
+## Delivery model
+
+Origin Data Lab combines automated anonymization with targeted Human QA to reduce manual workload while preserving delivery quality.
+
+## Responsible use
+
+Operators should confirm legal basis, data-handling requirements, retention policy, and final Human QA scope for each project.

README.md

@@ -1,3 +1,53 @@
----
-license: apache-2.0
----
+# dhaka11_anonymized_v58.mp4 Technical Report - V55-P0-FACE-ONLY
+
+**Origin Data Lab** — privacy processing and Human GT delivery for real-world video data.
+
+## Evaluation context
+
+- Engine version: `V55-P0-FACE-ONLY`
+- Pipeline version: `v55-p0-face-only`
+- Location: Not specified
+- Environment: Not specified
+
+## Processing metrics
+
+| Metric | Value |
+|---|---:|
+| Resolution | 1280 × 720 |
+| Processed frames | 2,661 |
+| Source FPS | 30.000 |
+| Processing FPS | 11.305 |
+| Elapsed seconds | 235.393 |
+| Detected faces | 10,634 |
+| Blurred faces | 10,634 |
+| Rejected regions | 0 |
+| Mean confidence | 0.6373 |
+
+### Processing speed
+
+![Processing speed](charts/fps_chart.png)
+
+### Detection activity
+
+![Detection activity](charts/detection_chart.png)
+
+### Confidence distribution
+
+![Confidence distribution](charts/confidence_chart.png)
+
+## Review and intended use
+
+Human QA is recommended before production delivery.
+
+This report documents an internal engineering evaluation. It is not an independently audited accuracy benchmark.
+
+## Known limitations
+
+- Very small or distant faces
+- Severe occlusion
+- Motion blur and compression artifacts
+- Difficult low-light scenes
+
+## Delivery model
+
+Origin Data Lab combines automated anonymization with targeted Human QA to reduce manual workload while preserving delivery quality.

dhaka11_anonymized_v58.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c3e5480a7d27913f5952c027330732f3ece0aace13aa35660b146aa317876ae5
+size 109263798

public_summary.json

@@ -0,0 +1,51 @@
+{
+  "schema_version": "odl-report-v2.1",
+  "generated_at_utc": "2026-07-25T07:12:40.502324+00:00",
+  "company": "Origin Data Lab",
+  "title": "Dhaka 1 Face Anonymization Test - V58",
+  "engine_version": "V55-P0-FACE-ONLY",
+  "pipeline_version": "v55-p0-face-only",
+  "location": "Dhaka, Bangladesh",
+  "environment": "Low-light dense urban traffic",
+  "input_video": "dhaka1 (1).mp4",
+  "anonymized_video": "dhaka11_anonymized_v58.mp4",
+  "metrics": {
+    "width": 1280,
+    "height": 720,
+    "source_fps": 30.0,
+    "source_frame_count": 2661,
+    "processed_frames": 2661,
+    "elapsed_seconds": 238.251,
+    "processing_fps": 11.169,
+    "detected_faces": 10634,
+    "blurred_faces": 10634,
+    "rejected_regions": 0,
+    "detection_rate_per_frame": 3.996242014280346,
+    "confidence_count": 10634,
+    "confidence_mean": 0.6373450762406763,
+    "confidence_min": 0.35005638003349304,
+    "confidence_max": 0.9466301798820496,
+    "region_types": {
+      "face_yunet": 10634
+    },
+    "settings": {
+      "face_confidence": 0.35,
+      "margin_ratio": 0.12,
+      "pixel_size": 34,
+      "blur_kernel": 61,
+      "shape": "ellipse",
+      "codec": "mp4v",
+      "max_frames": 0
+    }
+  },
+  "visual_review_score": null,
+  "reviewer_note": "Human QA is recommended for very small, occluded, low-light, or motion-blurred faces.",
+  "limitations": [
+    "Small faces",
+    "Occlusion",
+    "Motion blur",
+    "Low-light scenes"
+  ],
+  "human_review": "Recommended for production delivery",
+  "warnings": []
+}

report.html

@@ -0,0 +1,200 @@
+<!doctype html><html lang="en"><head><meta charset="utf-8"><meta name="viewport" content="width=device-width,initial-scale=1"><title>Dhaka 1 Face Anonymization Test - V58</title><style>
+body{margin:0;background:#0c111b;color:#edf2f7;font-family:Inter,Arial,sans-serif}main{max-width:1100px;margin:auto;padding:48px 24px}.brand{font-weight:800;letter-spacing:.08em;color:#7dd3fc}h1{font-size:42px;line-height:1.08;margin:14px 0}.sub{color:#aab7c8;max-width:800px}.hero,.panel,section{background:#141c29;border:1px solid #263448;border-radius:18px;padding:28px;margin:22px 0}.metrics{display:grid;grid-template-columns:repeat(auto-fit,minmax(160px,1fr));gap:12px;margin-top:24px}.metric{background:#0e1622;padding:16px;border-radius:12px}.metric span{display:block;color:#94a3b8;font-size:13px}.metric strong{display:block;font-size:22px;margin-top:7px}img{max-width:100%;height:auto;border-radius:12px;background:white}table{width:100%;border-collapse:collapse}td{padding:10px;border-bottom:1px solid #28364a}footer{color:#8290a3;margin-top:36px;font-size:13px}@media(max-width:640px){h1{font-size:32px}}
+</style><style id="v58-gallery-css">
+.showcase-lead {
+    margin: 0 0 18px;
+    color: #aab7c8;
+    line-height: 1.7;
+}
+
+.showcase-card {
+    background: #0e1622;
+    border: 1px solid #29384d;
+    border-radius: 16px;
+    padding: 16px;
+    margin-top: 18px;
+    break-inside: avoid;
+    page-break-inside: avoid;
+}
+
+.showcase-card h3 {
+    margin: 0 0 8px;
+    font-size: 20px;
+    color: #edf2f7;
+}
+
+.showcase-card p {
+    margin: 0 0 14px;
+    color: #94a3b8;
+    font-size: 14px;
+    line-height: 1.55;
+}
+
+.showcase-image {
+    display: block;
+    width: 100%;
+    border-radius: 12px;
+}
+
+.showcase-labels {
+    display: grid;
+    grid-template-columns: 1fr 1fr;
+    margin-bottom: 8px;
+    gap: 8px;
+}
+
+.showcase-label {
+    background: #182436;
+    border-radius: 8px;
+    padding: 8px 12px;
+    font-size: 12px;
+    font-weight: 700;
+    letter-spacing: .08em;
+    color: #cbd5e1;
+}
+
+.showcase-label:last-child {
+    color: #7dd3fc;
+}
+
+.showcase-grid {
+    display: grid;
+    grid-template-columns: repeat(2, minmax(0, 1fr));
+    gap: 16px;
+}
+
+.showcase-grid .showcase-card {
+    margin-top: 0;
+}
+
+.report-section {
+    break-inside: avoid;
+    page-break-inside: avoid;
+}
+
+@media(max-width:760px) {
+    .showcase-grid {
+        grid-template-columns: 1fr;
+    }
+}
+
+@media print {
+    section,
+    .showcase-card,
+    .report-section {
+        break-inside: avoid;
+        page-break-inside: avoid;
+    }
+
+    .showcase-grid {
+        grid-template-columns: 1fr 1fr;
+    }
+}
+</style>
+</head><body><main><div class="hero"><div class="brand">ORIGIN DATA LAB</div><h1>Dhaka 1 Face Anonymization Test - V58</h1><p class="sub">Automated privacy processing with targeted Human QA for real-world video data.</p><div class="metrics"><div class="metric"><span>Processed frames</span><strong>2,661</strong></div><div class="metric"><span>Processing FPS</span><strong>11.169</strong></div><div class="metric"><span>Detected faces</span><strong>10,634</strong></div><div class="metric"><span>Blurred faces</span><strong>10,634</strong></div><div class="metric"><span>Mean confidence</span><strong>0.6373</strong></div><div class="metric"><span>Engine</span><strong>V55-P0-FACE-ONLY</strong></div></div></div><section class="report-section" id="representative-examples">
+    <h2>Representative Before / After Examples</h2>
+
+    <p class="showcase-lead">
+        Representative frames were selected from different points in the video
+        to demonstrate face anonymization under low-light, dense urban traffic,
+        occlusion, motion blur, and mixed viewing distances.
+    </p>
+
+    <div class="showcase-card">
+        <h3>Primary demonstration — 00:08</h3>
+
+        <p>
+            Close-range riders and visible faces under low-light traffic conditions.
+            This frame provides the clearest visual comparison of the face-only
+            anonymization result.
+        </p>
+
+        <div class="showcase-labels">
+            <div class="showcase-label">BEFORE</div>
+            <div class="showcase-label">AFTER — FACE ANONYMIZATION</div>
+        </div>
+
+        <img
+            class="showcase-image"
+            src="report_assets/before_after_008s.jpg"
+            alt="Before and after face anonymization at 8 seconds">
+    </div>
+
+    <h3 style="margin-top:30px;">Additional representative frames</h3>
+
+    <div class="showcase-grid">
+
+        <div class="showcase-card">
+            <h3>00:03</h3>
+
+            <p>
+                Early dense-traffic scene with multiple visible road users.
+            </p>
+
+            <div class="showcase-labels">
+                <div class="showcase-label">BEFORE</div>
+                <div class="showcase-label">AFTER</div>
+            </div>
+
+            <img
+                class="showcase-image"
+                src="report_assets/before_after_003s.jpg"
+                alt="Before and after face anonymization at 3 seconds">
+        </div>
+
+        <div class="showcase-card">
+            <h3>00:08</h3>
+
+            <p>
+                Close-range faces demonstrating clear anonymization coverage.
+            </p>
+
+            <div class="showcase-labels">
+                <div class="showcase-label">BEFORE</div>
+                <div class="showcase-label">AFTER</div>
+            </div>
+
+            <img
+                class="showcase-image"
+                src="report_assets/before_after_008s.jpg"
+                alt="Before and after face anonymization at 8 seconds">
+        </div>
+
+        <div class="showcase-card">
+            <h3>00:30</h3>
+
+            <p>
+                Mid-video urban scene with mixed-distance faces and moving traffic.
+            </p>
+
+            <div class="showcase-labels">
+                <div class="showcase-label">BEFORE</div>
+                <div class="showcase-label">AFTER</div>
+            </div>
+
+            <img
+                class="showcase-image"
+                src="report_assets/before_after_030s.jpg"
+                alt="Before and after face anonymization at 30 seconds">
+        </div>
+
+        <div class="showcase-card">
+            <h3>01:01</h3>
+
+            <p>
+                Later low-light scene containing partial occlusion and complex traffic.
+            </p>
+
+            <div class="showcase-labels">
+                <div class="showcase-label">BEFORE</div>
+                <div class="showcase-label">AFTER</div>
+            </div>
+
+            <img
+                class="showcase-image"
+                src="report_assets/before_after_061s.jpg"
+                alt="Before and after face anonymization at 61 seconds">
+        </div>
+
+    </div>
+</section><section><h2>Processing speed</h2><img src="charts/fps_chart.png" alt="Processing speed"></section><section><h2>Detection activity</h2><img src="charts/detection_chart.png" alt="Detection activity"></section><section><h2>Confidence distribution</h2><img src="charts/confidence_chart.png" alt="Confidence distribution"></section><section><h2>Evaluation context</h2><table><tr><td>Pipeline</td><td>v55-p0-face-only</td></tr><tr><td>Location</td><td>Dhaka, Bangladesh</td></tr><tr><td>Environment</td><td>Low-light dense urban traffic</td></tr><tr><td>Reviewer note</td><td>Human QA is recommended for very small, occluded, low-light, or motion-blurred faces.</td></tr></table></section><section><h2>Responsible reporting</h2><p>This is an internal engineering evaluation, not an independently audited accuracy benchmark. Final production delivery should include Human QA for difficult scenes.</p></section><footer>Generated by Origin Data Lab Report Generator V2.1 Stable.</footer></main></body></html>

technical_report.pdf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:993c925853b46231d9c998a8bab01f743994446da826adfb7ed79041b298459a
+size 1054006