Hi Martin,
Thanks a lot for your response. I am tracking objects of various size and
number. Using various OpenCV algorithms including background subtraction
and optical flow, the objects I'm tracking end being K clouds of points,
where K varies throughout the sequence. I am a little new in this area,
but it seems like Kmeans would do a good job at clustering these clouds and
from my brief experience with kmeans in school I thought the common method
to determine K was to measure the degree of fit for several K values.
I will look into the paper you sent, and the other techniques you
mentioned. Thanks a lot for the help.
-Dave
On Thu, Apr 18, 2013 at 12:02 PM, Martin Fergie <[email protected]> wrote:
> Hi Dave,
>
> Can you shed a bit more light on what it is you are trying to cluster?
>
> You could try using cross validation along with one of the metrics below:
>
> http://scikit-learn.org/stable/modules/clustering.html#clustering-performance-evaluation
>
> However, this would require that you have a training set with ground truth
> labelings of segments, which often doesn't exist.
>
> Another option is to use Gaussian mixture models to estimate the number of
> clusters, the VBGMM model should select the number of clusters
> automatically. However, care must be taken as the fitted model has
> a tendency to over-generalise (i.e. too many clusters!).
>
> Another option is the model:
> http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=990138
>
> This fits a model with a large number of components and iteratively
> removes components, measuring the likelihood each time. You can then select
> the model which maximises the likelihood. This may work more effectively.
>
> These GMM techniques may not operate at real time, but you could try using
> them on some data gathered offline to gain an idea for how many clusters
> you should use for your real-time k-means algorithm.
>
> Hope it helps,
> Martin
>
>
> On 18 April 2013 15:41, David Reed <[email protected]> wrote:
>
>> Hi, I'm doing some realtime computer vision and have found the sklearn
>> implementation of KMeans is performing better at clustering my blobs
>> between frames.
>>
>> So I want to use sklearn, but unfortunately I don't know the number of
>> blobs, K, and was wondering if sklearn can tell me the fit of the model. I
>> was looking through the documentation but I wasn't sure if what I was
>> looking for was there.
>>
>> Also, if any one has any experience clustering blobs in CV and thinks
>> this is the wrong direction to go in, please let me know.
>>
>> Thanks a lot for the help.
>>
>> Dave
>>
>>
>> ------------------------------------------------------------------------------
>> Precog is a next-generation analytics platform capable of advanced
>> analytics on semi-structured data. The platform includes APIs for building
>> apps and a phenomenal toolset for data science. Developers can use
>> our toolset for easy data analysis & visualization. Get a free account!
>> http://www2.precog.com/precogplatform/slashdotnewsletter
>> _______________________________________________
>> Scikit-learn-general mailing list
>> [email protected]
>> https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
>>
>>
>
>
> ------------------------------------------------------------------------------
> Precog is a next-generation analytics platform capable of advanced
> analytics on semi-structured data. The platform includes APIs for building
> apps and a phenomenal toolset for data science. Developers can use
> our toolset for easy data analysis & visualization. Get a free account!
> http://www2.precog.com/precogplatform/slashdotnewsletter
> _______________________________________________
> Scikit-learn-general mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
>
>
------------------------------------------------------------------------------
Precog is a next-generation analytics platform capable of advanced
analytics on semi-structured data. The platform includes APIs for building
apps and a phenomenal toolset for data science. Developers can use
our toolset for easy data analysis & visualization. Get a free account!
http://www2.precog.com/precogplatform/slashdotnewsletter
_______________________________________________
Scikit-learn-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
