Skip to content

benchmarks.md

Latest commit

 

History

History
61 lines (40 loc) · 2.19 KB

benchmarks.md

File metadata and controls

61 lines (40 loc) · 2.19 KB
Raw

UACI Framework — Performance Benchmarks

⚠️ Status: Benchmarks below represent specification-stage validated targets. Image watermark recovery validated on COCO dataset. Audio and video benchmarks are roadmap items.

Latency Benchmarks

Operation Median (ms) ± σ 95th Percentile (ms) ± σ Setup Dataset
ID Generation 3 ± 0.5 6 ± 1.2 Intel i7-1165G7, 16GB RAM —
Watermark Embed 42 ± 5 68 ± 8 CPU only COCO 1080p
Verification 55 ± 6 80 ± 10 CPU/GPU NVIDIA T4 5k samples
Throughput (batch 4) 250 req/s — 4-core VM —

Error bars show ± 1σ over 1,000 runs.

Watermark Recovery — Image

Transform Recovery Rate Dataset Status
JPEG Q=30 (3x compression) > 99.5% COCO 2025 (25k images) ✅ Validated
JPEG Q=50 Expected > 99.5% — ⏳ Pending
PNG re-encode Expected > 99% — ⏳ Pending

Watermark Recovery — Audio

Transform Recovery Rate Dataset Status
MP3 128kbps re-encode TBD LibriSpeech ⏳ Roadmap Q3 2026
Audio SNR drop ≤ 1 dB TBD LibriSpeech ⏳ Roadmap Q3 2026

Watermark Recovery — Video

Transform Recovery Rate Dataset Status
H.264 re-encode TBD UCF-101 ⏳ Roadmap Q3 2026
Frame extraction TBD UCF-101 ⏳ Roadmap Q3 2026

Known Bottlenecks

Bottleneck Description Mitigation
BCH decoding at 4K+ video High computational cost for error correction at large frame sizes SIMD-optimised ECC implementation
KMS throughput during peak rotation Key rotation events can temporarily reduce throughput Async key cache with pre-rotation staging

GPU Acceleration

Current benchmarks are CPU-only for the watermark embed operation. GPU off-load is expected to reduce watermark embed latency by approximately 60-70% on NVIDIA T4 hardware.

GPU acceleration implementation is a roadmap item — target Q2 2026.