Cull Me uP Lab · live readouts

The numbers that matter.

Real shoots, real seconds — no synthetic sets. The detail lives in each room.

RAW decode · Sony ARW

×5.4

faster than Apple ImageIO

Turbo26.1s

Classic139.8s

Splash Dogs · 409 photos · 26.13 GB, same Mac

Cull Me uP RATE

6,190/h

peak — a fast, decisive cull

RAW throughput

Decode throughput bars, Turbo versus classic

10–16img/s

Sony ARW · Turbo (classic does 3–4)

Inside the Lab · how Nala learns

We show the engine learning.

No black box. You rate, it calibrates to your taste, and every detection model is benchmarked in the open — the transparency you'd want from a tool that touches your selects.

Training — star-rating a frame with zone-aware sharpness boxes (face, centre, thirds)

01 · NOTATION

You set the standard.

Rate frames ★1–5. Zone-aware sharpness — face, centre, edges — teaches the engine what sharp means to you.

See the AI page →

Corpus — per-folder calibration, before/after scores and correlation ρ

02 · CALIBRAGE

It calibrates to your taste.

A pass over your folder tunes the engine to your own ratings — correlation, not guesswork. ρ −0.26 → −0.02.

See the AI page →

Analyse modèles — model benchmark table: time, precision, sharpness stats

03 · ANALYSE

Every model, measured.

Detection models benchmarked head-to-head — speed, precision, sharpness min/avg/max. Full transparency.

See Algorithms →