Command Overview

Synthesizing Interaction snapshots

To synthesize an interaction snapshot, use v synth in the command line. Options are described below:

Option	Description
i=[text]	Text description for the interaction snapshot to generate. The text description is parsed by the specified parser into set of verb and nouns. By default, the current interaction set (from UI) is used.
parserType=[parserType]	Which parser to use: label (default) or nlp
classifierType=[classifierType]	Which classifier to use (default to currently selected classfifier from pulldown in UI)
`selectModel=[random	first]`
`selectSkel=[Average	Sample
fromScan	Indicates that the current scan should be used as a constraint
skipModelPlacement	Indicates that model placement should be skipped
predictSegmentLabels	Whether segment labels should be predicted
keepExisting	Whether to keep existing model instances or not. By default, scene is cleared each time synth is performed.
naive	In naive mode, PIGs are not used for placement.
compose	For composite actions, use composition to form the interaction snapshot (instead of taking probabilities directly from composite PIGs.
retarget	For verbs applied to new nouns, try to use existing verb PIG and generalized it for new noun.
backoff	Do a backoff, where we try to use the composite PIGs we have seen as much as possible. If not available, then backoff to composition and retargeting.
debug	For debug mode
log	Log debug output scene and debug visualizations. Use o to specify the output directory. If o is not specified ouput is stored in $workDir/isynth/test. The output consists of the scene (scene.json) and the log file (log.txt). If debug mode is set, debug visualization is also stored.)
o=[outputDir]	Output directory of where the interaction snapshot should be saved. Unless outputDir is an absolute path, the output is stored relative to $workDir/isynth. Output is only generated if log is specified.

Synthesizing unconstrained snapshots

v synth
v synth i=sit_chair-look_tv
v synth "i=he is sitting in a chair" parserType=nlp

Synthesizing constrained snapshots

v synth fromScan

It is also possible to synthesize constrained snapshots using the v place command. Options to v place are given below.

Option	Description
skelRange	Use current skeleton range for placement
predictAction	Whether possible action should be predicted
predictSegs	Whether segment labels should be predicted
debug	For debug mode

Saving generated snapshots

To save to ply, use v writePLY models skeleton.

To save to json, use v saveScene <filename>.

Other commands

After synthesizing a scene, you can clear it by pressing Shift-I.

Use v testSnaps <nSnapsPerInteraction> <filename> to batch generate interaction snapshots. Use <filename>=known to generate interaction snapshots for all known interactions.

Generating interaction set and PiGraph model statistics

There are several commands to dump out aggregated interaction set statistics. Output is placed in the configured workDir, typically scenegrok\bin\work. When creating PIGs, the graphml for the pigs are placed in $workDir/pigs/graphs

Command	Description
v dumpPIGFeats	Dumps csv files with the PIG features in $workDir/pigs/feats. Use all to iterate over all interaction sets, otherwise PIG features for only the current iteraction set is dumped.
v dumpCatProbs	Dumps csv files with joint to category probabilities and joint to label probabilities in $workDir/pigs/jc. Use all to iterate over all interaction sets, otherwise statistics for only the current iteraction set is dumped.
v saveInteractionStats	Dumps csv files with aggregated statistics for all the interaction sets in $workDir/pigs/stats. Includes aggregated noun to joint group probabilities, noun to scene type probabilities, verb to joint probabilities, etc.