ISP Analysis Platform · Part of the 10xEngineers Imaging Suite
- ISP-Native · HDR-Aware · Pipeline-Integrated
Cut your tuning loop from hours to minutes
LumaIQ replaces fragmented scripts, disconnected tools, and manual YAML edits with one integrated IQ analysis workbench — built specifically for ISP algorithm engineers.
3
Integrated apps
LumaIQ · RawIQ · BatchIQ
24
ColorChecker patches
CIE76 · CIE94 · CIEDE2000 ΔE
13+
RAW formats supported
.nef .arw .cr3 .dng + sensor dumps
At a glance
Fast read for engineers evaluating a switch
Built on PyQt6, NumPy, SciPy, and OpenCV. Runs natively on Windows and Linux. No cloud dependency, no subscription lock-in.
PyQt6
Windows
Linux
rawpy
NumPy
SciPy
OpenCV
Pipeline
YAML-native
Full collapsible per-stage editor
HDR support
Native
Dual-exposure · LTM · Tone map
Batch runs
GUI + CLI
CSV-driven overnight regression
Reports
Auto-generated
Self-contained HTML, base64 charts
Presets
Tuning Vault
Named preset CRUD across sessions
UI thread
Never blocked
All runs in background QThreads
The problem it solves
Every iteration costs minutes.
That adds up to days.
ISP tuning today is a fragmented loop. Each tool switch is a source of error and delay. LumaIQ collapses the entire cycle into one environment.
-
Current workflow — fragmented
1
Capture RAW frame on hardware
Trigger sensor, extract dump manually
2
Edit YAML config by hand
800-line config, one parameter at a time
3
Run Python ISP script
Scripts “kind of work” — only one person understands them
4
Export PNG → open Imatest
Switch tools, lose context, wait for load
5
Copy numbers into Excel
Shared sheet that’s broken twice this quarter
6
Repeat from step 2
Three weeks of this per tuning cycle
-
Current workflow — fragmented
1
Load RAW file
Drop your .nef, .arw, .raw, .dng — LumaIQ handles it
2
Load your YAML config
Collapsible per-stage editor, in-app — no text editor needed
3
Run ISP pipeline
Background thread — UI stays responsive throughout
4
Analyze output instantly
MTF, SNR, ΔE, dynamic range — all computed in the same tool
5
Generate & share report
One-click HTML export — base64-embedded, zero dependencies
6
Adjust and repeat — instantly
Change parameter → see result. No tool switches, no waiting
Features
Built for real pipelines, not photography labs
Every feature exists because an ISP engineer needed it during tuning — not because a product manager needed a checkbox.
⚡
ISP Pipeline Integration
Loads any YAML-based ISP config with a full collapsible per-stage parameter editor. Run the pipeline directly from the UI in a background thread — output captured for downstream analysis.
YAML-native
Live edit
Background run
🌗
HDR / Linear Mode Toggle
HDR-only stages — exposure selection, alpha blending, slope correction, dual-channel NR, tone mapping — auto-show and hide. The tool knows your mode and surfaces what’s relevant.
Dual-exposure
Auto-stage reveal
🎯
Color Accuracy — ColorChecker
Automatic ColorChecker prompt on image load. Quad-corner boundary selector extracts all 24 patches and computes CIE76, CIE94, and CIEDE2000 ΔE per patch with OK / Warn / Fail grading.
CIEDE2000 ΔE
24 patches
Severity grade
📦
Tuning Vault
Named preset CRUD for every calculator — save, load, browse, and share parameter sets across sessions. Stored at ~/.lumiq/vault/. Your tuning knowledge survives the next onboarding.
Cross-session
Shareable
📝
HTML Report Generator
Self-contained dark-themed HTML export with base64-embedded charts. Single file, zero external dependencies — share with stakeholders without worrying about broken asset paths or tool installations.
One file
Dark theme
base64 charts
🔍
Planckian AWB
Full Hernandez-Andres Planckian locus, sensor curve calibration, block chromaticity voting, and CCT estimation from multi-illuminant captures — not a slider, an actual algorithm.
Planckian locus
CCT estimation
Analysis modules
Every IQ metric that matters.
Nothing you don't need.
LumaIQ covers the complete measurement chain — from raw pixel statistics through color science to perceptual quality scores.
| Module | Key Metrics | Category |
|---|---|---|
| Sharpness / MTF | ISO 12233 slanted-edge SFR · Laplacian variance · Sobel edge strength · per-channel MTF curves | Spatial |
| Noise & SNR | Per-channel σ (R/G/B) · CIELAB luma/chroma noise (σL* σa* σb* σC*) · SNR vs luminance curve · 1D noise power spectrum · spatial noise heatmap · zone breakdown | Noise |
| Dynamic Range | EV estimate from flat-block SNR analysis · Ansel Adams 5-zone system (Deep Shadow → Specular) · zone overlay maps · histogram | Exposure |
| HDR Tone Mapping | Reinhard global TMO · Durand bilateral LTM · side-by-side before/after preview · live tone curve overlay · hardware LUT compare with PSNR and diff heatmap | HDR |
| Color Accuracy | CIE76 · CIE94 · CIEDE2000 ΔE per patch · per-patch severity grading (OK / Warn / Fail) | Color |
| White Balance | Per-channel WB calculation · metrics · hue/sat fix · Planckian locus · CCT estimation | Color |
| Exposure & Sensitivity | Section-wise histograms on sensor frames · HDR pair detection · LDR best-exposure selector | Exposure |
| Lens Calibration | OpenCV chessboard pipeline · corner detection · camera calibration · undistortion map generation · persistent calibration JSON | Optics |
| Contrast & Gamma | Entropy · local contrast · halo score · gamma estimation · live stats bar · LDCI analyze and calculate | Tonal |
HDR integration
Close the loop with LumaIQ
Once your HDR ISP pipeline is tuned, the next question is always: how good is the output, really? LumaIQ plugs directly into your existing HDR ISP backend — reading your YAML configs, running your pipeline stages, and measuring results across every critical IQ dimension.
🔗
Zero configuration gap
Reads the same YAML config your HDR ISP uses. No translation, no adapter — the parameter you edit is the parameter that runs.
📐
Quantified, not subjective
CIEDE2000 ΔE, MTF curves, and SNR-vs-luminance tell you whether a tuning decision is genuinely better — not just visually different.
⚙️
HDR-aware from the start
Dual-exposure pair matching, exposure selection, slope correction, and alpha blending — all HDR stages visible and editable in one place.
HOW IT WORKS
Separable H/V polyphase filtering
Two stages — horizontal then vertical — each implemented as a multi-tap, multi-phase FIR. Coefficients are selected per phase (fractional position), enabling high-quality interpolation for arbitrary ratios.
-
Multi-tap FIR
Configurable taps (4–12) to balance quality vs resources.
-
64 phases
Phase banks represent sub-pixel positions for consistent detail.
-
Anti-aliasing built-in
Downscales remain stable without extra AA blocks.
-
Low latency pipeline
Streaming architecture for real-time paths.
Pipeline view
Conceptual block flow
AXI4-Stream Input
RGB / YCbCr, 8/10/12-bit (cfg dependent)
Horizontal & Vertical Polyphase FIR
multi-tap filtering + Line buffering + phase-aligned interpolation
AXI4-Stream Output
Low-latency scaled stream
Bypass mode
1:1 pass-through when scaling is not needed.
Coefficient banks
Static or dynamic loading via AXI4-Lite.
Note: exact buffering depends on format, taps, and throughput configuration.
FORMAT & INTEGRATION
Video-friendly I/O and control
Designed for streaming pipelines while exposing the knobs you actually need: scaling parameters and coefficient banks.
- Video format support
Config dependent
- RGB 4:4:4
- YCbCr 4:2:0
- YCbCr / YC 4:2:2
- 8 / 10 / 12 bpc
- Interfaces
- AXI4-Stream for input/output video streaming
- AXI4-Lite for programming and runtime control
- Programmable coefficients
Support for dynamically or statically loaded coefficient banks — select responses for different product modes.
Configuration knobs
Typical product controls
Scaling factors
Arbitrary H / V ratios
Independent per axis
Filter complexity
4–12 taps
Quality ↔ resources
Phase resolution
64 phases
Sub-pixel alignment
Memory strategy
BRAM / URAM options
Pixel-rate dependent
For higher pixel rates, the scaler can be optimized to utilize URAMs instead of BRAMs (configuration dependent).
Technology-independent soft IP Core for FPGA, ASIC and SoC devices (license: EULA).
APPLICATIONS
Where polyphase scaling pays off
If your product needs stable detail, controlled aliasing, and predictable latency, polyphase filtering becomes a quiet enabler — especially in premium video and imaging.
-
Broadcast imaging
Real-time 4K scaling with low latency and high fidelity across formats — built for professional pipelines.
-
Video conferencing
Clean resizes across mixed sources (camera + content) with consistent sharpness and fewer artifacts.
-
Media walls
Ultra-smooth 4K scaling, precise alignment, and dependable visuals for multi-display systems.
-
VR headsets
Low latency and crisp imagery — critical when scaling content into display-native panels.
-
Medical imaging
High-fidelity 4K scaling that preserves fine diagnostic detail, supports low-latency visualization, and enables high-quality interpolation in medical display pipelines.
RESOURCE UTILIZATION
Measured with OOC flow (Vivado)
Utilization results are captured using Vivado out-of-context (OOC) flow. Surrounding design impacts placement and routing, so treat these as guidance. At higher pixel rates, memory can be optimized toward URAM instead of BRAM (configuration dependent).
Example configuration matrix
Replace with your measured data
| FPGA | Scaler Configuration | Resource Utilization | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Max Width | Max Height | Input Format | Output Format | Pixel Bit Width | Taps | Pixels per clock | LUT | FF | BRAMs | URAMs | DSP | ||
| 1 | zu5ev | 3840 | 2160 | 420 | 420 | 8 | 6 | 1 | 3528 | 2442 | 13 | 0 | 44 |
| 2 | zu5ev | 3840 | 2160 | 420 | 420 | 8 | 12 | 1 | 7552 | 4116 | 22 | 0 | 80 |
| 3 | zu5ev | 3840 | 2160 | 420 | 420 | 10 | 12 | 1 | 10052 | 5524 | 36 | 0 | 80 |
| 5 | zu5ev | 3840 | 2160 | 420 | 420 | 12 | 6 | 1 | 5290 | 3384 | 21.5 | 0 | 44 |
| 6 | zu5ev | 3840 | 2160 | 420 | 420 | 12 | 12 | 1 | 11348 | 6886 | 36.5 | 0 | 80 |
| 7 | zu5ev | 4096 | 2160 | 444 | 444 | 8 | 6 | 4 | 14141 | 9425 | 0 | 18 | 198 |
| 8 | zu5ev | 4096 | 2160 | 444 | 444 | 10 | 6 | 4 | 15182 | 10452 | 0 | 18 | 198 |
| 9 | zu5ev | 4096 | 2160 | 444 | 444 | 12 | 6 | 4 | 22206 | 13902 | 0 | 18 | 198 |
Final numbers depend on format, taps, pixel rate, memory strategy, and surrounding pipeline.
NEXT STEP
Want to see it in your pipeline?
Share your formats, resolutions, and target device. We’ll map a configuration (taps/phases/memory strategy) and provide integration package details.
-
Integration-ready interfaces
AXI4-Stream + AXI4-Lite -
Low-latency streaming
Built for real-time systems
-
Quality knobs that matter
Taps, phases, coefficients -
Portability
Structured RTL for porting
Contact
By requesting info, you agree to be contacted about this IP.