[Topic-models] Topic models application to video data

Ayan Acharya masterayan at gmail.com
Thu Jul 15 23:48:50 EDT 2010

Hi All,

  I am looking for a 3 level LDA model (a level added on top of basic LDA
like the case with multiple corpus). Is there any code available that can be
readily extended to implement this model?

Ayan Acharya

(Graduate Student,
Track: Communication, Networks and Systems,
Electrical and Computer Engineering,
University of Texas at Austin,
Austin, Texas, USA)

On Thu, Jul 15, 2010 at 6:09 AM, Eric Wang <exw.duke at gmail.com> wrote:

> Toyin,
> There's been some work in this area.  This paper by Wang, Ma, and Grimson
> proposes two models, based on LDA and HDP, respectively, to address activity
> modeling. I found it well written, informative, and thorough.
> http://citeseerx.ist.psu.edu/viewdoc/download?doi=
> And here's a more recent paper by Wang and Carin that proposes a temporally
> and spatially dependent topic model for activity perception (this is my
> work)
> http://umiacs.umd.edu/~jbg/nips_tm_workshop/10.pdf<http://umiacs.umd.edu/%7Ejbg/nips_tm_workshop/10.pdf>
> Both papers use quantized Lucas-Kanade optical flow fields to describe
> "visual words" that then are analyzed by the proposed topic models.
> Hope this helps.
> Eric
> On Thu, Jul 15, 2010 at 5:24 AM, Toyin Popoola <toyin_net at yahoo.com>wrote:
>> Hello All,
>> I joined this list today after reading David Blei's paper on Topic models.
>> I want to apply this concept to video, for modelling different types of
>> human activity.
>> I have been able to extract spatio-temporal features in terms of 'visual
>> words' which represents each video clip as a distribution over these code
>> words.
>> Now I want to apply HDP to this dataset to build a generative model that
>> captures the structure in each class of activity.
>> My goal is for the model to be able to corectly classify an unseen data
>> into one of the identified classes, and to detect an abnormal activity.
>> Does anyone have suggestions on how I can go about this?
>> I will be grateful.
>> Toyin
>> _______________________________________________
>> Topic-models mailing list
>> Topic-models at lists.cs.princeton.edu
>> https://lists.cs.princeton.edu/mailman/listinfo/topic-models
> _______________________________________________
> Topic-models mailing list
> Topic-models at lists.cs.princeton.edu
> https://lists.cs.princeton.edu/mailman/listinfo/topic-models
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.cs.princeton.edu/pipermail/topic-models/attachments/20100715/40d3b7d0/attachment.html>

More information about the Topic-models mailing list