A while ago, I was so amazed about the image segmentation results using Gaussian Mixture Models (GMMs) because GMM gives pretty good results on normal/natural images. There are some results on my previous post. Of course, GMM is not the best for this job, but hey look at its speed and easiness to implement–it’s pretty good in that sense. However, one problem with GMM is that we need to pick the number of components. In general, the more component numbers we assume, the better log-likelihood it would be for GMM. In that case, we would simply send the number of components to infinity, right? Well…but there is nothing good come out of that because the segment would not be so meaningful–in fact, we overfit the data, which is bad.
Therefore, Bayesian Information Criteria (BIC) is introduced as a cost function composing of 2 terms; 1) minus of log-likelihood and 2) model complexity. Please see my old post. You will see that BIC prefers model that gives good result while the complexity remains small. In other words, the model whose BIC is smallest is the winner. Simple as that. Here is the MATLAB code. Below are some results from sweeping the number of components from 2 to 10. Unfortunately, the results are not what I (and maybe other audiences) desire or expect. As a human, my attention just focuses on skier, snow, sky/cloud and perhaps in the worst case, the shadows, so the suitable number of components should be 3-4. Instead, the BIC assigns 9-component model the winner which is far from I expected. So, Can I say that the straightforward BIC might not be a good model for image segmentation, in particular, for human perception? Well…give GMM-BIC a break– I think this is too early to blame BIC because I haven’t use other more sophisticated features like texture, shape, color histogram which might improve results from using GMM-BIC. The question is what are the suitable features and the number of components that makes the segmentation results using GMM-BIC similar to human perception? MATLAB code is made available here.
If you want to put UF identities (e.g. logo, template, color, font, etc.) on your presentation slides, businesscard, website, etc., you can find everything in the “UF Identity” – http://identity.ufl.edu/ and more specifically “signature system” – http://identity.ufl.edu/signatureSystem/
Here are some database for image segmentation/classification
- NEC Animal Dataset [link]
- Microsoft Research Object Classification Dataset [link]
- Microsoft Research Webcam Foreground/Background Dataset [link]
- Motion-based Segmentation and Recognition Dataset (Cam Vid) [link]
- The PASCAL Object Recognition Database Collection [link] — this page links to other image databases.
- UCI image database [link]
- Berkeley Image Database [link]
- Caltech 101 [link]
- LabelMe [link]
- collections of popular image database from Princeton Vision Lab [link]
Conference and Workshops
- Click here to see the list
- TPAMI: The IEEE Transactions on Pattern Analysis and Machine Intelligence
- JMLR: Journal of Machine Learning Research
- JAIR: Journal of Artificial Intelligence Research
- Machine Learning (Springer)
Websites and Blogs
- Videolectures.net: This is a great website for machine learning enthusiasms. Tons of lectures and tutorials, all in video format. Updated regularily. You can find all interesting lectures from a lot of conferences, workshops and MLSS here.
- NIPS: Here you can find tons of papers and tutorial videos.
Also I found a journal/conference list in Bob Wang’s homepage very helpful.
How to download .flv from YouTube
If you are using the browser Firefox, you can easily download .flv file from the website by simply using an add-on “DownloadHelper”. Please refer to the link below
There are so many softwares that can download the video file from YouTube. However, I prefer using the firefox add-on, DownloadHelper since it’s really easy.
How to convert FLV file to AVI
Of course, AVI is like a crucial material for any kind of video/media file. Once we can convert a file to AVI, then we can go from there to all types of file so easily. There are a lot of freeware of this sort. The one I’m using right now is Pazera Free FLV to AVI converter.
This link also contains tons of other media file converters.
How to convert/rip FLV to MP3
Again there are a lot of freewares on the internet. The one I’m using is http://www.flv2mp3.com/
I think the website works pretty well, so far as good. The website not only converts the file from URL, but also can convert the file on your machine as well. The website will upload the video from the source, then convert the FLV to MP3. When finish, you will get a small window pop up and ask you to save the MP3 file on your machine. However, the file does not have extension, don’t panic, you just put the extension .mp3 afterward and every thing will work just fine!
Hold on to the left-click and drag your mouse around the video!
I found this video from the “immersive media” website http://demos.immersivemedia.com
It looks like the company is the one behind the scene of Google StreetView technolgy — http://dvice.com/archives/2007/05/dodeca_2360_camera_on_top_of_a.php
Check this out: http://www.immersivemedia.com
Today I had a chance to attend a very good seminar about Art of Communications: Marketing you ideas. The presenter was Megan Gales, the director of communications. Her presentation was so great; the slides look stylish and persuasive. The slides were made by Keynote from mac. Please check this website for the power of this keynote, it’s amazing! I got entertained through the whole presentation.
I’m just wondering if can I make Powerpoint look very stylish like that of keynote?
Anyway, this is the first time I heard about the unit that can help UF students and faculty to create a great communications skill. They even have graphic consultant! Ummm…UF provides such a great facility for its community.