- Version 1.0.1-272-ga0864af

mrcal-show-residuals - Visualize calibration residuals in an imager

```
$ mrcal-show-residuals --vectorfield left.cameramodel
... a plot pops up showing the vector field of residuals for this camera
```

This tool supports several different modes, selected by the commandline arguments. Exactly one of these mode options must be given:

```
--vectorfield Visualize the optimized residuals as a vector field. Each
vector runs from the observed chessboard corner to its
prediction at the optimal solution.
--magnitudes Visualize the optimized residual magnitudes as color-coded
points. Similar to --vectorfield, but instead of a vector, each
residual is plotted as a colored circle, coded by the MAGNITUDE
of the error. This is usually more legible.
--directions Visualize the optimized residual directions as color-coded
points. Similar to --vectorfield, but instead of a vector, each
residual is plotted as a colored circle, coded by the DIRECTION
of the error. This is very useful in detecting biases caused by
a poorly-fitting lens model: these show up as clusters of
similar color, instead of a random distribution.
--regional Visualize the optimized residuals, broken up by region. The imager
of a camera is subdivided into bins. The residual statistics are
then computed for each bin separately. We can then clearly see
areas of insufficient data (observation counts will be low). And
we can clearly see lens-model-induced biases (non-zero mean) and
we can see heteroscedasticity (uneven standard deviation). The
mrcal-calibrate-cameras tool uses these metrics to construct a
valid-intrinsics region for the models it computes. This serves as
a quick/dirty method of modeling projection reliability, which can
be used even if projection uncertainty cannot be computed.
--histogram Visualize the distribution of the optimized residuals. We display
a histogram of residuals and overlay it with an idealized
gaussian distribution. If the optimization was successful, and
if observed_pixel_uncertainty represented the input errors
faithfully AND there was no overfitting, the two would line up.
All of these rarely occur together, and usually the predicted
distribution is significantly less certain than the observed
distribution: there's usually a strong overfitting effect.
```

```
model Camera model that contains the optimization_inputs
that describe the solve. The displayed observations
may come from ANY of the cameras in the solve, not
necessarily the one given by this model
```

```
-h, --help show this help message and exit
--vectorfield Visualize the optimized residuals as a vector field
--magnitudes Visualize the optimized residual magnitudes as color-
coded points
--directions Visualize the optimized residual directions as color-
coded points
--regional Visualize the optimized residuals, broken up by region
--histogram Visualize the distribution of the optimized residuals
--valid-intrinsics-region
If given, I overlay the valid-intrinsics regions onto
the plot. Applies to all the modes except --histogram
--gridn GRIDN GRIDN How densely we should bin the imager. By default we
use a 20x14 grid of bins. Applies only if --regional
--histogram-this-camera
If given, we show the histogram for residuals for THIS
camera only. Otherwise (by default) we display the
residuals for all the cameras in this solve. Applies
only if --histogram
--binwidth BINWIDTH The width of binds used for the histogram. Defaults to
0.02 pixels. Applies only if --histogram
--vectorscale VECTORSCALE
Scale all the vectors by this factor. Useful to
improve legibility if the vectors are too small to
see. Applies only if --vectorfield
--extratitle EXTRATITLE
Extra title string for the plot
--hardcopy HARDCOPY Write the output to disk, instead of making an
interactive plot
--terminal TERMINAL gnuplotlib terminal. The default is good almost
always, so most people don't need this option
--set SET Extra 'set' directives to gnuplotlib. Can be given
multiple times
--unset UNSET Extra 'unset' directives to gnuplotlib. Can be given
multiple times
```

https://www.github.com/dkogan/mrcal

Dima Kogan, `<dima@secretsauce.net>`

Copyright (c) 2017-2020 California Institute of Technology ("Caltech"). U.S. Government sponsorship acknowledged. All rights reserved.

Licensed under the Apache License, Version 2.0 (the "License"); You may obtain a copy of the License at

` http://www.apache.org/licenses/LICENSE-2.0`