A 100% local, cross-platform camera RAW optical and sensor metrology workbench (Web App + Tauri v2 Desktop). Powered by Rust + WebAssembly, all RAW files are decoded and analyzed strictly in local memory (zero uploads) with zero-latency 1:1 pixel inspection.
- 🌐 Online Workbench: https://h-void.github.io/collima/?lang=en (English · 简体中文 · 繁體中文 · 日本語)
- 📦 Desktop Downloads: GitHub Releases (Windows x64/ARM64 · macOS Apple Silicon/Intel · Linux x64/ARM64)
macOS tip: Builds use ad-hoc signing. If Gatekeeper blocks first launch, move
collima.appto/Applicationsand runxattr -cr /Applications/collima.apponce.
-
Optical Tilt Inspector — 9-zone synchronized 1:1 nearest-neighbor loupe (
100% / 200% / 400%), 5×5 full-frame contour heatmap, and polar tilt compass. Measures sharpness directly on pre-demosaic$G_1 / G_2$ planes (plus Star HFR) to separate normal wide-open field curvature from mount/element tilt without vignetting or lateral CA bias. -
Sensor Noise Lab — Per-channel (
$R, G_1, G_2, B$ ) black-level drift, read noise ($\sigma _ {\text{ADU}}$ & ISO 100 normalized), Engineering Dynamic Range (EV), automatic Dual Native ISO (HCG) switch detection, and CSV/JSON export. -
Built-in 14-Bit Synthetic Suite & Reactive 4-Locale UI — Try instantly without camera RAWs via in-memory synthesized 14-bit
.DNGcalibration frames; switch languages anytime without re-decoding 60 MP RAW buffers.
Optical & Sensor Metrology Formulas
-
Black-Level-Calibrated Normalized Tenengrad (evaluated independently on raw Bayer
$G_1$ and$G_2$ sub-planes so optical vignetting and fixed black-level pedestals do not underestimate corner sharpness):$$\displaystyle S _ {G_k} = \frac{1}{N \cdot (\mu _ {G_k} - \text{BL} _ {G_k})^2} \sum _ {(x,y)} \max\left(0, G_x^2 + G_y^2 - \tau _ {\text{noise}}^2\right)$$ -
Tilt vs. Curvature Decoupling: Weighted Least Squares (WLS) fits a 1st-order tilt plane alongside a 2nd-order radially symmetric falloff term.
-
Read Noise & Engineering Dynamic Range (single-pass
$O(N)$ exact 65,536-binu16histogram with iterative$5\sigma$ hot-pixel rejection):$$\displaystyle \text{DR} _ {\text{EV}} = \log_2 \left( \frac{\text{WhiteLevel} - \text{BlackLevel}}{\sigma _ {\text{ADU}}} \right)$$
| Engine | Cameras & Compression Codecs |
|---|---|
flashraw 0.3 |
~40 RAW formats: Sony .ARW / .ARQ (incl. ARW 6.0 X-OCN / LLVC3 Compressed HQ on α7 V / α7R VI & Lossless), Canon .CR3 / .CR2, classic Nikon .NEF, Adobe .DNG (incl. DNG 1.7 JPEG XL), Fujifilm .RAF (incl. X-Trans), Lumix/Leica .RW2 / .RWL, OM .ORF, Pentax .PEF, Hasselblad .3FR, Phase One .IIQ. |
ticoraw 0.2 |
Nikon High Efficiency .NEF (HE & HE*) clean-room bit-exact JPEG XS wavelet decoder for Z9, Z8, Zf, Z6III, Z5II, Z50II, and ZR (with EXIF fused via flashraw). |
- Lens Tilt (1 RAW frame): Shoot a flat textured surface perpendicular to the optical axis (distant building facade, brick wall filling all four corners, or a zenith starfield) wide open, at base ISO, on a tripod (IBIS off, EFCS/electronic shutter), focused precisely at the center (
MC). - Sensor Noise & Dual Native ISO (Multi-ISO Dark RAWs): Put the body cap on and cover the viewfinder in a dark room. Use
M+MF, Long Exposure NR off,1/1000 s–1/4000 sshutter speed, and shoot one uncompressed/lossless RAW per ISO step (e.g., ISO 100–6400).
Local Build & Development
Requires Rust >= 1.85 (with wasm32-unknown-unknown), wasm-pack, Node.js >= 20, and optional Tauri v2 prerequisites for desktop packaging:
cargo test --workspace # Run Rust optical & sensor unit tests
cd web && npm install
npm run build:wasm && npm run dev # Start browser workbench
npm run build # Build production static web bundle
npm run tauri:build # Package native desktop installers (.exe/.msi, .dmg, .deb/.rpm/.AppImage)collima's optical tilt/curvature solver, pre-demosaic dual-green Tenengrad, 65,536-bin read-noise & Dual Native ISO engine, 14-bit .DNG synthetic calibrator, and 9-zone 1:1 workbench are original implementations. RAW decoding and runtime infrastructure build upon:
- @ybouane /
flashraw(MIT OR Apache-2.0) — Pure-Rust RAW decoder (~40 formats incl. SonyARW 6.0X-OCN, CanonCR3,DNG 1.7) & color matrices. - @lukaj99 /
ticoraw(MIT OR Apache-2.0) — Pure-Rust clean-room decoder for Nikon High Efficiency.NEF(HE/HE*). - Ecosystem:
jxl-oxide,zune-image,wasm-bindgen,tauri,react, andlucide— see THIRD-PARTY-NOTICES.md.
Dual-licensed under MIT or Apache-2.0, at your option.