Training

Once some observations are stored in the database, models can be built from them.

Config File

Training, just like recognition, requires a configuration file but it usually only contains one cell that defines a pipeline that reads data from the database and computes a model. Then again, nothing prevents you to train several models at the same time by executing several pipelines.

Use

Just run the training.py script in /apps. It requires a configuration file through the -c option. Some of the options in there can be overriden through the command line for convenience. Change the following parameters to your needs:

  • the object_ids should be the list of object ids you want to train on. If you want, you can also use object_names,

    that are more human readable. object_ids is of the form [“6b3de86feee4e840280b4caad90003fb”] but there are two special options: if it is “all”, then all models are recomputed; if it is “missing”, only the missing models are computed.

The training script can be run as follow:

rosrun object_recognition_core training \
-c `rospack find object_recognition_tod`/conf/training.ork \
--visualize

You can choose whatever configuration file; a few are provided in object_recognition_server/conf.

A typical command line session might look like:

% apps/training -c config_training.txt --commit
Prepare G2O: processing image 65/65
Performing full BA:
iteration= 0     chi2= 168324165740673896546304.000000   time= 39.2803   cumTime= 39.2803        lambda= 154861.907021 edges= 64563     schur= 1
Persisted

Command Line Interface

No module named PySide.QtCore
usage: training [-h] [-c CONFIG_FILE] [--visualize] [--commit]

optional arguments:
  -h, --help            show this help message and exit
  -c CONFIG_FILE, --config_file CONFIG_FILE
                        Config file
  --visualize           If set and the pipeline supports it, it will display
                        some windows with temporary results
  --commit              Commit the data to the database.