Re:Computer Vision in the Real World – Why SIGGRAPH Probably...
by acgourley on Tuesday September 17, 2013 @04:42PM
(#6401821)
OP here - Just wanted to say that I believe pipelined architecture is a symptom of iterative research and is not necessarily good. I think the more direct solution avoids pipelining in favor of a domain specific hack.
[ Parent
| on HN ]
Re:Computer Vision in the Real World – Why SIGGRAPH Probably...
by yetanotherphd on Tuesday September 17, 2013 @06:20PM
(#6402304)
Thanks for making that point. It still seems to me that most domain specific solutions will still make use of many ideas from the CV literature. E.g. looking at the last sections of your video it seems like you settled on a histogram based approach, although I may be wrong.
Re:Computer Vision in the Real World – Why SIGGRAPH Probably...
by acgourley on Wednesday September 18, 2013 @02:43PM
(#6407145)
I think you get a bag of tricks as you work on this stuff, for sure. But I think they are more like the tricks artisans gather, share and re-use than fundamental academic truths.
Re:Computer Vision in the Real World – Why SIGGRAPH Probably...
by joe_the_user on Tuesday September 17, 2013 @08:45PM
(#6402791)
Agreed,
I think the main thing about pipelining, modularity and such is that researchers want to produce something that someone else can use.
Domain specific hacks do seem like the most effective in practice, they more or less by definition can't be reused or extended, however.
Which can be sad from the perspective of making progress over time.