Overview of the 2012 challenge

This year's challenge will focus on recognizing an object's:

  1. Identity, and
  2. 6DOF (degree-of-freedom) pose with respect to a reference calibration pattern.

There will be several scense of such objects both attached to calibration rigs (when recognition and pose is required) and free form (where only recognition is required). Objects are to be recognized by type (e.g. wine glass, coke can) and not by class.

Objects will be:

  1. Rigid,
  2. not shiny,
  3. not transparent,
  4. small enough that one could lift them with one hand

These include, but are not limited to, consumer goods, auto parts, and geometric objects.


Scenes will be of increasing difficulty where contestants will have to identify:

  1. Single objects and their pose.
  2. Multiple objects with occlusion.
  3. Multiple objects and their pose with occlusion.

Contestants can use any 2D or 3D sensor(s) that they want, they can use any computing equipment that they want including communicating with external "cloud computing". Some objects to be learned for the contest will be made known prior to the competition, the others (the majority) will have to be learned at the conference prior to the competition day.

Please contact us at solutionsinperception@gmail.com for questions.

Rules

The solutions in perception committee's judgement define the results of the contest. The committee can add, delete or modify the rules at any time.

Prior to the conference:

  • As stated above: Contestants can use any 2D or 3D sensor(s) that they want; they can use any computing equipment that they want including communicating with external "cloud computing".
  • Some objects to be learned for the contest will be made known prior to the competition.
  • Object capture and test fixtures designs will be made available to contestants.
  • Contestants will be required to pass a set of milestones (being able to post scores to our contest server, passing some preliminary recognition tests, etc) prior to acceptance in the competition. Committee decisions on acceptance are final.

At the conference:

  • Contestants can bring any 2D or 3D sensor(s) and computing equipment and/or communicate with an external "cloud" for training and testing.
  • Contestants will be given access to the actual objects to be recognized as soon as possible upon arrival at the conference.
  • Contestants should bring training fixtures and other necessary equipment by which to learn the objects in the challenge.
  • Contestants may start training as soon as they get the objects. They will have one full day in the competition room for further training and testing on preliminary fixtured scenes (which will be changed on the competition day)
  • On the training day, contestants will display conference style posters that describe their techniques in the competition room.

During the competition:

  • Contestants can use any 2D or 3D sensor(s) and computing equipment that they can bring to, or communicate with, at the conference.
  • Contestants can use any robotic device or hold the scensor(s) by hand for collecting the data.
  • There can be no human assistance with the data from the sensors for object recognition or object pose identification.
  • Contestants will have a fixed duration of time in which to move the sensors around the object scenes prior to reporting the results to the competition servers.

After the competition:

  • Prizes will be awarded to the winners, if any, in an exponentially rising fashion the closer the scores are to perfection. Actual amounts and schedule of awards will be posted later.
  • There will be an awards ceremony which contestants are encouraged to attend.
  • Contest results and method descriptions belong to the committee and will be posted publically afterwards.
  • It is encouraged that contestants make their code available to others for testing and comparison in future times.