[Dprglist] AI controlled camera at soccer game

David Anderson davida at smu.edu
Wed Nov 4 14:52:19 PST 2020


Hi Karim,

Hmmm..  "quality of the image database" seems to suggest there is some 
art involved.  (I like that).

I'm still not clear as to what is meant by non-positive examples.  Does 
each image in the training set need to be marked in advance somehow as 
to it's relevance?   I don't really know how this works.

thanks!
David


On 11/2/20 7:02 PM, Karim Virani wrote:
> My AI thinks I'm a mop. COVID hair.
>
> Pretty much all visual object detection/classification neural nets 
> require a healthy number of non-positive examples. This looks like a 
> fail in exactly that - the body shape always being under the "ball" 
> should have lowered the probability if they had bald soccer players in 
> the general set. Though I guess photos of bald soccer players with 
> extremely round heads heading soccer balls could get a little fuzzy...
>
> Anyhow, in our limited experience, the quality of the image database 
> seems to matter as much or more than the classification algorithm.
>
> On Mon, Nov 2, 2020 at 6:13 PM David Anderson via DPRGlist 
> <dprglist at lists.dprg.org <mailto:dprglist at lists.dprg.org>> wrote:
>
>     As long as the players don't make the same mistake you'll probably
>     be ok...
>
>     My favorite part is when the ball gets kicked straight toward the
>     lineman.  The camera moves very quickly and smoothly to track the
>     path but ends up locked on the lineman's head when a player
>     intercepts the ball.
>
>     So for those of you more familiar with neural nets, is there a way
>     to train it up on what not to recognize? Sort of an inverse
>     training?  Say, if the soccer ball is attached to a neck and
>     shoulders, it's probablhy not a soccer ball?   Can you show it a
>     bunch of pictures of things that are _not_ soccer balls to improve
>     it's recognition?
>
>     cheers,
>     David
>
>
>
>     On 11/2/20 5:38 PM, Doug Paradis wrote:
>>     Same thing happens to me when I play lineman.
>>
>>     On Mon, Nov 2, 2020 at 1:32 PM David Anderson via DPRGlist
>>     <dprglist at lists.dprg.org <mailto:dprglist at lists.dprg.org>> wrote:
>>
>>         Evidently using an AI controlled camera that's supposed to
>>         follow the
>>         ball but gets confused by the lineman's bald head.
>>
>>         https://www.youtube.com/watch?v=9zoJP2FkpgU
>>
>>         In fairness, in that lighting, his head does look a lot like
>>         the ball.
>>
>>         cheers!
>>
>>         dpa
>>
>>
>>         _______________________________________________
>>         DPRGlist mailing list
>>         DPRGlist at lists.dprg.org <mailto:DPRGlist at lists.dprg.org>
>>         http://lists.dprg.org/listinfo.cgi/dprglist-dprg.org
>>
>
>     _______________________________________________
>     DPRGlist mailing list
>     DPRGlist at lists.dprg.org <mailto:DPRGlist at lists.dprg.org>
>     http://lists.dprg.org/listinfo.cgi/dprglist-dprg.org
>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.dprg.org/pipermail/dprglist-dprg.org/attachments/20201104/8b6cc3eb/attachment.html>


More information about the DPRGlist mailing list