ALS Flat Calibration A/B Report

Generated from /home/deufrai/Desktop/flat_tests

Verdict: the flat-calibrated run is healthy. It processed the same 14 lights with no warnings or errors, applied cached dark and flat masters correctly, preserved alignment stability, and measurably improved field and color uniformity.

Note: image measurements are made on ALS final processed PNG outputs, after autostretch/display processing. A stricter scientific validation should also use linear stack outputs.

Run Summary

RunFramesProfileDarkFlatWarnings / ErrorsSearched ratiosAccepted ratios
No flat141yesno0 / 00.10.1
Flat141yesyes0 / 00.10.1
Flat division overhead~212 ms/framemean in calibrated run
Accepted match ratio0.1all aligned frames in both runs
Warnings / errors0 / 0both logs
Final image size6248 x 417616-bit RGB PNG

Visual Comparison

A/B comparison and difference map

Top-left: no-flat result. Top-right: flat-calibrated result. Bottom-left: signed flat-minus-no-flat difference map, where green means brighter after flat calibration and magenta means darker.

Timing

MetricUnitNo flat nNo flat medianNo flat meanNo flat minNo flat maxFlat nFlat medianFlat meanFlat minFlat max
Preprocessms144767.6725181.4134333.3686203.659144912.3324912.1964607.1775333.645
Stackms148657.6467958.5103.3629229.913148773.0908011.1373.3049115.200
Full frame processings1421.64621.70913.03424.5951422.41521.57613.88422.911
Dark subtractionms1477.61478.36374.32086.5301478.22679.55175.18090.505
Flat divisionms0----14210.908211.777196.561234.252

Per-frame Log Comparison

Light suffixNo flat matchesFlat matchesNo flat sigmaFlat sigmaNo flat pre msFlat pre msNo flat stack msFlat stack msNo flat full sFlat full s
222832_-9.8C_0002.fit4402.15333.63.43.313.03413.884
223333_-9.8C_0003.fit49494333.44607.27565.67558.120.34820.530
223634_-9.8C_0004.fit48485832.84653.17477.17733.121.68620.863
224004_-9.6C_0005.fit4848413474441696644349.14747.78527.78640.021.30722.067
224307_-9.8C_0006.fit4848368912436764874403.04791.18549.68331.621.39021.607
224648_-9.6C_0007.fit4848360637735510574543.74910.58699.18948.621.58022.429
224950_-9.8C_0008.fit4848315039031258784510.04933.28616.19115.221.55022.805
225330_-9.7C_0009.fit4848226049322680384807.54849.98709.88887.022.22822.461
225632_-9.9C_0010.fit4848204335720473724727.94869.98459.78686.821.60622.497
225949_-9.8C_0011.fit4747207685520812966150.24914.18882.08594.623.63021.989
230250_-9.8C_0012.fit3647228351320332566092.04980.38902.08905.423.51222.401
230600_-9.8C_0013.fit4747204547720132096100.14914.88895.78983.723.60222.911
230904_-9.8C_0014.fit4747199673319967196084.55041.69229.98859.424.59522.757
231212_-9.8C_0015.fit4747176900017646796203.75223.68901.58909.023.85522.869

Final Image Luminance

RunMeanStd devp1p5Medianp95p99
No flat13197.85717.181268707118112338140085
Flat13200.35583.788019269118382305139753

The calibrated image lifts the dark tail while keeping global brightness essentially unchanged. That is consistent with vignetting/response correction rather than simple brightening.

Field Uniformity

RegionNo flat median / centerFlat median / centerNo flat medianFlat medianFlat - noflat median
center201.0001.00012944.312154.0-795.7
tl200.8610.97211145.711814.0698.0
tr200.6810.7728815.79378.7560.3
bl200.9041.03211697.712542.7891.7
br200.6930.8038966.09760.3809.7
top100.8390.91910856.011166.3163.3
bottom100.8570.95111098.311558.0536.7
left100.9231.04611949.712716.0799.7
right100.6870.7888893.39575.0600.3

Tile background uniformity

RunMean tile backgroundMinMaxRangeCVp10p90
No flat11486.18181.325278.717097.30.21368948.013220.4
Flat11509.38857.024628.315771.30.19759408.712904.2

Color Uniformity

MetricNo flatFlatNo flat rangeFlat range
R0.26510.245920790.019357.0
G0.16910.162113935.013141.0
B0.20970.186416694.014975.0
R/G spatial std0.08090.0701
B/G spatial std0.03670.0223

Interpretation

The flat-calibrated image darkens the brighter central field and lifts the darker outer field. The strongest measured changes are in the corners and side edges, especially the right edge and both right corners.

Alignment remains stable: all aligned frames accepted at the restored progressive ratio 0.1. Match counts are similar or better with flat calibration; one no-flat frame had only 36 matches where the flat run had 47.

Sigma clipping remains comparable between runs. The calibrated run has a slightly lower average clipped sample count, but the difference is small enough to read as stable behavior rather than a behavior shift.

Source Files