1

Suppose you have a ground plane and can use a stereo vision system to detect things that are possibly separate objects.

Suppose also your robot or agent can attempt to pick up and move these objects around in real-time.

Is there any current system in computer vision that allows new objects to be learned in real-time?

nbro
  • 39,006
  • 12
  • 98
  • 176
FourierFlux
  • 783
  • 1
  • 4
  • 14
  • By "real-time" do you mean "incrementally", i.e. learn new classes incrementally as they arrive, and not necessarily strictly "online" and "real-time" (i.e. fast)? – nbro Aug 15 '20 at 13:25
  • Both possibly, basically not require retraining of the whole system. – FourierFlux Aug 15 '20 at 14:22

1 Answers1

1

You are probably looking for incremental learning (sometimes known as lifelong learning) techniques, i.e. machine learning techniques that attempt to address the catastrophic forgetting effect of neural networks when trained incrementally, i.e. as new classes or data are added to the original training data.

There are different techniques and some of them store (or compress) the old data in order to fully or partially re-train the neural network with the new classes or data. However, note that this is a relatively new area of research and significant progress still needs to be made to produce serious tools. If you are specifically interested in incremental class learning, maybe have a look at this paper Class-incremental Learning via Deep Model Consolidation (2020).

nbro
  • 39,006
  • 12
  • 98
  • 176