[4eyes] Papers for Friday

[Brynjar Gretarsson]

The size of the two paper nodes represents the length of the paper (number
of characters, figures not considered). The size of the topic nodes
represents the topic relevance for the set of documents (in this case there
are only two documents). And yes they are normalized so that there is a
maximum node size.

For those interested in how the topics were mined, I simply used an LDA
algorithm which takes in the set of documents and the number of topics that
I want to find. It should be noted that LDA works better the bigger the set
of documents is. I have not read the papers and am certainly not an expert
in the field that they discuss, so perhaps somebody else could figure out
what each of the ten topics stands for. Usually you should be able to
represent each set of words with a concept or an entity.

For example t4 seems to be about model reconstruction (using kinect), while
t7 seems to be about tracking. Note that this is a list of the top 10 words
in order, so the first words are more important although the set as a whole
should represent the concept.

Thanks,
Brynjar

On Thu, Nov 3, 2011 at 9:02 AM, Byungkyu Kang <bk.kang at me.com> wrote:

> This is cool, Brynjar!
>
> By the way, does the size of each node represent the number of occurrence
> in each paper? and the size of a node is calibrated(normalized) for the
> comfort of viewer?
>
> -Jay
>
> On 2011. 11. 3., at 오전 12:22, Brynjar Gretarsson wrote:
>
> Hi all,
>
> Attached is an image of the two papers and their associated topics. For
> this I ran the topic modeling algorithm with T=10 topics. Of the 10 topics,
> 3 are mostly associated with the ISMAR paper (cyan), 2 are mostly
> associated with the UIST paper (dark blue), and 5 are associated with both
> (light blue). It should be noted that the edges drawn depend on a threshold
> that we select, but this image still indicates that there is a topical
> difference between the two papers. The relative topic node positions
> reflect topic similarity.
>
> For those interested these are the 10 topics detected (each topic shown as
> a list of top 10 words and top 2 words are shown in the image):
> [t1] surface measurement global depth tsdf frame pixel sdf space current
> [t2] map distance pose prediction function values association projective
> previous signed
> [t3] vision dense real mapping research proceeding ieee robotic images
> scanning
> [t4] kinect model reconstruction interaction kinectfusion rgb user multi
> interactive geometry
> [t5] real icp physical figure particles dynamic static surface pipeline
> virtual
> [t6] ray volume voxel vertex normal grid gpu transform thread truncated
> [t7] tracking depth data live representation camera perform scenes
> rendering small
> [t8] sensor frame mapping loop scene slam model result processing
> resolution
> [t9] reconstruction figure motion quality large frames data shown non
> ability
> [t10] scene object foreground user background moving touch reconstructed
> physic core
>
>
> I can show you the interactive visualization on Friday if there is
> interest and time.
>
> Thanks,
> Brynjar
>
> On Wed, Nov 2, 2011 at 6:50 PM, Domagoj <domagoj at cs.ucsb.edu> wrote:
>
>>
>> UIST paper: http://research.microsoft.com/**apps/pubs/default.aspx?id=**
>> 155416 <http://research.microsoft.com/apps/pubs/default.aspx?id=155416>
>>
>> ISMAR paper: http://research.microsoft.com/**apps/pubs/default.aspx?id=**
>> 155378 <http://research.microsoft.com/apps/pubs/default.aspx?id=155378>
>>
>>
>>
>> On 2011-11-02 17:17, Brynjar Gretarsson wrote:
>>
>>> Hi,
>>>
>>> Does anyone have the two papers that we talked about yesterday and will
>>> be presented on Friday. If I can get the text from both papers I might
>>> try to create a topic based visualization of the two papers to see if
>>> they are topically very similar and show it on Friday.
>>>
>>> Thanks,
>>> Brynjar
>>>
>>>
>>> ______________________________**_________________
>>> Ilab-users mailing list
>>> Ilab-users at lists.cs.ucsb.edu
>>> https://lists.cs.ucsb.edu/**mailman/listinfo/ilab-users<https://lists.cs.ucsb.edu/mailman/listinfo/ilab-users>
>>>
>> ______________________________**_________________
>> Ilab-users mailing list
>> Ilab-users at lists.cs.ucsb.edu
>> https://lists.cs.ucsb.edu/**mailman/listinfo/ilab-users<https://lists.cs.ucsb.edu/mailman/listinfo/ilab-users>
>>
>
> <KinectPapers.gif>_______________________________________________
>
> Ilab-users mailing list
> Ilab-users at lists.cs.ucsb.edu
> https://lists.cs.ucsb.edu/mailman/listinfo/ilab-users
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cs.ucsb.edu/pipermail/ilab-users/attachments/20111103/3bf83941/attachment-0001.html>