Skip to content

Updates to IMU-based activity detection

Jonathan Jones requested to merge activity-detection into master

Updates to project structure:

  • deleted libduplo
  • deleted metawear
  • renamed libstats -> assemblystats
  • renamed libprimesense -> primesense_datastream
  • renamed pymetawear_logging -> metaweardevice
  • renamed sweep_inference_params -> sweep_inference_param
  • moved StatsPipeline from compute_stats.py into its own file

Updates to annotation interface:

  • incorporated BlockAssembly object to parse and display annotated state
  • added display box for notes
  • unbind u/i/j/k keys from frame navigation (use arrow keys only from now on)
  • cosmetic updates
  • refactoring, style improvements

Updates to data collection interface:

  • style improvements

Updates to BlockAssembly:

  • visualization methods take file extension args, return image filename
  • style improvements

Updates to compute_stats.py:

  • move sub-experiment code into its own function
  • change exception handling if script is run from ipython
  • refactored, deleted redundant code
  • style improvements

Updates to definitions.py:

  • added imu/video sample rates
  • added block activity label definitions
  • style improvements
  • moved metadata dtypes from DuploCorpus
  • moved event2str method (duplicated in compute_stats/libduplo)

Updates to DuploCorpus:

  • added method to read CSV-format notes
  • all definitions are now imported from definitions.py
  • style improvements

Updates to InferencePipeline:

  • added subclass labels for supervised training of activity/likelihood model
  • separated data processing for rgb/depth and accel/gyro

Updates to inferstate.py:

  • read/write/cache notes
  • added subclass labels
  • separated data processing

Updates to labels.py:

  • moved actions2str from compute_stats
  • renamed extractBlockActivitySeqs -> extractBlockActionSeqs
  • new implementation of extractBlockActivitySeq: parses notes files in format "grasp/release [, ]*"

Updates to metaweardevice.py:

  • style improvements

Updates to metrics.py:

  • renamed 'edge' metrics to 'avg' metrics

Updates to models.py:

  • likelihood model takes optional subclass label sequences
  • HMM can be initialized with extra counts during parameter estimation (see Hmm.fit())

Updates to primesense_datastream.py:

  • style improvements

Updates to rawdata.py:

  • preprocessSignalSeqs can handle NoneType args

Updates to utils.py:

  • Added window length, stride length computation functions

Merge request reports