Skip to main content
Version: 9.0

A Typical MetriCal Workflow

Users can find workflow scripts and sample datasets in the MetriCal Sensor Calibration Utilities repo on GitLab:
https://gitlab.com/tangram-vision/platform/metrical/-/tree/main/examples

1: Initialize MetriCal​

A plex is a virtual snapshot of your system's calibration at a point in time. Init creates a raw input plex by inferring it from the data it is given. We'll optimize this plex, and this data, using Calibrate later.

metrical init \
  --topic-to-model *:opencv_radtan                    \ # Assign all of our models
  --preset-device RealSense435:[ir_one,ir_two,color]  \ # Set any pre-configured devices
  $DATA                                               \ # Provide our data
  $INIT_PLEX                                            # Our output: a fresh plex!

More on...​

2: Calibrate the Initial Plex​

The actual calibration! This refines the plex, giving you accurate calibation values. We'll visualize this process using Rerun by passing the --render flag.

metrical calibrate            \
  --render                    \ # Render the calibration process using Rerun
  --output-json $OUTPUT       \ # Write the results to $OUTPUT
  $DATA                       \ # Our input calibration dataset
  $INIT_PLEX                  \ # Our fresh plex from Step 1
  $OBJ                        \ # Our object space file describing fiducials and targets

More on...​

3: Shape the Calibration Results​

This shapes the calibration output into something more useful for our system. In this case, we want to create a stereo rectification for our system.

metrical shape       \
  $OUTPUT            \ # Use the calibration output we derived from Step 2
  $OUTPUT_DIR        \ # Write the results to $OUTPUT_DIR
  stereo-lut         \ # Create a pair of stereo rectification lookup tables
  --dominant ir_one  \ # The dominant eye in this stereo pair
  --secondary ir_two \ # The secondary eye in this stereo pair

And there you have it! In three easy (and scriptable) lines, we went from raw data to a calibrated, rectified stereo pair.

More on...​