Discussion/Problems

6. Discussion / Problems

Some Problems with visual Routines

Appears that some complex computations are performed in order to perceive what appears as simple properties and relations. However, our knowledge of how how our visual system / brain performs these seemingly simple computations.is still very incomplete! How are visual routines implemented in the brain? and how can these routines be simulated by humans to imitate our visual system? The following are the main problems the author associates with visual routines:

Elemental operations - what are they?
Integration - how are they arranged into routines?
Control - how are they selected, which ones are selected?
Assembly and Compilation: how are visual routines constructed to meetspecific goals and how are the routines controlled? How are new routines generated, compiled and stored?
Storage problem: Are routines stored in memory rather than being constructed each time they are need to meet some specific goal? Routines stroed in memory are certainly more efficient.

6.1 My Personal Views:

Have visual routines been implemented? Yes, there has been plenty of research / work performed with respect to visual routines especially at the MIT's AI lab (for example, see [4]). The following links provide further information and include implementations of visual routines and various publications:

MIT AI Lab's COG Project.
Visual routines for Autonomous Driving by Garbis Salgian.
Understanding People Pointing: The Perseus System by Roger E. Kahn and Michael J. Swain.
Visual Routines a summary by Meredith Frost.
Home page of Michael Patrick Johnson, MIT Media Laboratory. Has links to several publications related to visual routines including his Master's thesis titled "Evolved Visual Routines".
Home page of Ian Horswill. A former member of the AI lab at MIT who has done extensive work related to visual routines, (see [4] for an example).
Finally, there is a tremendous amount of information related to visual routines. Try a search using google - there are many links available!

There is probably a lot more to the human visual system than what can be explained using visual routines - plenty of problems must be solved! Furthermore, we are very far from constructing anything which may remotely imitate the human system!

6.2 Comments Regarding the Paper:

Paper was very extensive and thorough providing a great deal of details. Despite being a rather large document, it was very enjoyable to read, simple to follow and certainly very interesting.

Finally, despite the time which has passed since the paper first appeared (almost 18 years), visual routines have certainly not been "shot down" . There have been additions and modifications however, visual routines still remain a very large topic of interest even today.