Image Parsing: Unifying Segmentation, Detection, and Recognition

Recently I have posted about the connection between object category detection and semantic image segmentation. While delving into this problem more deeply, I have found the paper denoted in the title of this post.

The ICCV 2003 paper by Tu et al. was among the three Marr prize winners. And it is really a prominent piece of work! In the introduction to the special Marr prize IJCV issue, Bill Triggs featured the paper as one that "would have been particularly pleasing to David Marr", because of "its bold attack on the central problem of perceptual organization and whole scene understanding". Here is the journal version of the paper.

The authors combined discriminative and generative models, which resulted to the unified framework for image parsing. For each image the corresponding parsing tree could be found. The root of the tree represents the whole scene. On the next level, nodes represent semantic fragments, such as human faces, text, or textured regions. The leaves of the tree correspond to the pixels of the image.

The approach differs drastically from the vanilla CRF framework in the way that the structure of the parsing tree is dynamic while the CRF structure remains constant and just interaction models between the pairs of sites may change. The goal is to obtain the tree that maximizes the posterior probability given the image. The directed search in the space of valid trees is performed by means of Markov chain Monte-Carlo (MCMC). The possible tree changes like split and merge of regions, varying the border between regions, are defined. Such changes are described in terms of learnable Markov chain transition kernels.

What they actually did is they effectively combined top-down and bottom-up approaches. In a nutshell, the bottom-up (discriminative) method generates the hypotheses of pixel labels and object positions using the local neighbourhood, and the top-down generative model is then built making use of those hypotheses. The latest guarantees the consistency of the output and the optimum of its posterior probability.

Okay, what does it mean for the issue of detection and segmentation convergence? Because the shapes of objects are determined implicitly during the recognition, and stuff regions are detected by their colour and texture, the problem of semantic image segmentation is actually solved. Moreover, multi-scale semantic segmentation is performed during the parsing. Therefore, image parsing seems to be the most general formulation of image recognition problem.

The nodes of a parsing tree, which belong to the same depth level, are not connected (right, because it is a tree). This means that spatial interactions could not be modelled directly. A totally different approach was introduced in another Marr prize winning paper by Desai et al [2009]. They also use bottom-up tests to generate hypotheses on the object locations and then use CRF over those location to take spatial context into account. They model different kinds of inter-class and intra-class interactions. Thus, they ensure that the objects in the image (described by their bounding boxes) are arranged correctly, e.g. a cup is likely to be on a table (spatial arrangement), while a train could not be close to a ship (mutual exclusion).

It seems that the two papers exploit the two different forms of information (aggregation vs. spatial interactions), which are mutually redundant to a certain extent, but do not completely exhaust each other. It seems that the group of researchers who will successfully combine those approaches will receive some 201x Marr prize.

Can a Statin Neutralize the Cardiovascular Risk of Unhealthy Dietary Choices?

The title of this post is the exact title of a recent editorial in the American Journal of Cardiology (1). Investigators calculated the "risk for cardiovascular disease associated with the total fat and trans fat content of fast foods", and compared it to the "risk decrease provided by daily statin consumption". Here's what they found:

The risk reduction associated with the daily consumption of most statins, with the exception of pravastatin, is more powerful than the risk increase caused by the daily extra fat intake associated with a 7-oz hamburger (Quarter Pounder®) with cheese and a small milkshake. In conclusion, statin therapy can neutralize the cardiovascular risk caused by harmful diet choices.

Routine accessibility of statins in establishments providing unhealthy food might be a rational modern means to offset the cardiovascular risk. Fast food outlets already offer free condiments to supplement meals. A free statin-containing accompaniment would offer cardiovascular benefits, opposite to the effects of equally available salt, sugar, and high-fat condiments. Although no substitute for systematic lifestyle improvements, including healthy diet, regular exercise, weight loss, and smoking cessation, complimentary statin packets would add, at little cost, 1 positive choice to a panoply of negative ones.

Wow. Later in the editorial, they recommend "a new and protective packet, “MacStatin,” which could be sprinkled onto a Quarter Pounder or into a milkshake." I'm not making this up!

I can't be sure, but I think there's a pretty good chance the authors were being facetious in this editorial, in which case I think a) it's hilarious, b) most people aren't going to get the joke. If they are joking, the editorial is designed to shine a light on the sad state of mainstream preventive healthcare. Rather than trying to educate people and change the deadly industrial food system, which is at the root of a constellation of health problems, many people think it's acceptable to partially correct one health risk by tinkering with the human metabolism using drugs. To be fair, most people aren't willing to change their diet and lifestyle habits (and perhaps for some it's even too late), so frustrated physicians prescribe drugs to mitigate the risk. I accept that. But if our society is really committed to its own health and well-being, we'll remove the artificial incentives that favor industrial food, and educate children from a young age on how to eat well.

I think one of the main challenges we face is that our current system is immensely lucrative for powerful financial interests. Industrial agriculture lines the pockets of a few large farmers and executives (while smaller farmers go broke and get bought out), industrial food processing concentrates profit among a handful of mega-manufacturers, and then people who are made ill by the resulting food spend an exorbitant amount of money on increasingly sophisticated (and expensive) healthcare. It's a system that effectively milks US citizens for a huge amount of money, and keeps the economy rolling at the expense of the average person's well-being. All of these groups have powerful lobbies that ensure the continuity of the current system. Litigation isn't the main reason our healthcare is so expensive in the US; high levels of chronic disease, expensive new technology, a "kitchen sink" treatment approach, and inefficient private companies are the real reasons.

If the editorial is serious, there are so many things wrong with it I don't even know where to begin. Here are a few problems:

1. They assume the risk of heart attack conveyed by eating fast food is due to its total and trans fat content, which is simplistic. To support that supposition, they cite one study: the Health Professionals Follow-up Study (2). This is one of the best diet-health observational studies conducted to date. The authors of the editorial appear not to have read the study carefully, because it found no association between total or saturated fat intake and heart attack risk, when adjusted for confounding variables. The number they quoted (relative risk = 1.23) was before adjustment for fiber intake (relative risk = 1.02 after adjustment), and in any case, it was not statistically significant even before adjustment. How did that get past peer review? Answer: reviewers aren't critical of hypotheses they like.
   
2. Statins mostly work in middle-aged men, and reduce the risk of heart attack by about one quarter. The authors excluded several recent unsupportive trials from their analysis. Dr. Michel de Lorgeril reviewed these trials recently (3). For these reasons, adding a statin to fast food would probably have a negligible effect on the heart attack risk of the general population.
   
3. "Statins rarely cause negative side effects." BS. Of the half dozen people I know who have gone on statins, all of them have had some kind of negative side effect, two of them unpleasant enough that they discontinued treatment against their doctor's wishes. Several of them who remained on statins are unlikely to benefit because of their demographic, yet they remain on statins on their doctors' advice.
   
4. Industrial food is probably the main contributor to heart attack risk. Cultures that don't eat industrial food are almost totally free of heart attacks, as demonstrated by a variety of high-quality studies (4, 5, 6, 7, 8, 9). No drug can replicate that, not even close.
   

I have an alternative proposal. Rather than giving people statins along with their Big Mac, why don't we change the incentive structure that artificially favors the Big Mac, french fries and soft drink? If it weren't for corn, soybean and wheat subsidies, fast food wouldn't be so cheap. Neither would any other processed food. Fresh, whole food would be price competitive with industrial food, particularly if we applied the grain subsidies to more wholesome foods. Grass-fed beef and dairy would cost the same as grain-fed. I'm no economist, so I don't know how realistic this really is. However, my central point still stands: we can change the incentive structure so that it no longer artificially favors industrial food. That will require that the American public get fed up and finally butt heads with special interest groups.

Open peer review done right

There's been a lot of talk recently on the idea of open peer review, open source science, and so on. arXiv was sort of intended to serve this role, but at least in my subfield it's uncommon to put papers on there before they've been accepted by a "proper" conference/journal. In general we use it for open archival purposes, but not for initial peer review.

Image by Network Osaka

However, I recently saw a really nice use of open peer review. Vinay Deolalikar of HP Labs sent a manuscript to a few of his colleagues which was a proof of P ≠ NP. This is one of the most fundamental unsolved problems in CS, in fact it's such an important problem that The Clay Mathematics Institute offers a $1 million prize to anyone who can solve it. I won't burden you with the details of the problem or proof because that's not what this post is about, and will instead refer you to this excellent summary in New Scientist if you are interested.

So Deolalikar sent this proof to his colleagues, and Greg Baker (with permission) posted the manuscript on his blog. Somehow the blog post got Slashdotted and next thing you know everyone was talking about it. What I find really remarkable was how in many places there was a nice, civil discussion among the theory types about the strengths and weaknesses of the proof. (You can read about this in Richard Lipton's blog starting here).

What I especially loved was the peanut gallery. People were just thrilled to be watching the action. Even people with no background on the topic were offering support and encouragement. Here are some comments from Lipton's blog:

 Go Vinay!

I agree with the spirit of this posting. I am not an expert in complexity theory so I am unable to provide any feedback (not even elementary). However, I like it that there are researchers like Vinay who are not afraid of tackling the hardest problems. I am anxious to know what are the consequences of Vinay’s work.

-Anon@ 11:23pm

And these two really made me say, "awwww" (boldface is mine):

Professor Lipton,

Your accessible and well-written writings motivate me to continue studying mathematics. I develop software for a living; to stay sharp, I try to study ‘real’ proofs in my spare time. Complexity is an amazing field… I cannot wait to see what we’ll learn about computation (and its relation to physics, and to everything else).

Thanks for the inspiration!

-Robert W 

“P.S. I hope these discussions are helpful to the community at large.”

This member of the community at large can’t understand a word you say, but is nevertheless fascinated by every new post and comment. Seeing the review process unfold in public has rekindled my long-dormant interest in mathematics. I intend to register for a class this fall and (who knows?) perhaps pursue that Ph.D. after all these years.

Don’t infer from the paucity of experts who can contribute to the public conversation that you might as well confer entirely in private; on the contrary, public discussion is an immense service to the community. Thank you.

-Fritz Juhnke

I applaud Lipton for cultivating a positive exchange of ideas and place for civil discussion and making the review public for all to watch. Even though things got heated at times, in general the discourse was constructive and positive. Top complexity theoreticians came together to help Deolalikar out, for no reason other than they think it's a cool problem and they're nice people, and I think that's just fantastic.

ALTERNATIF HERBAL UNTUK ATASI LIMFOMA....

Limfoma atau kelenjar getah bening dicirikan adanya benjolan yg biasa terdapat di bagian leher atau bagian-bagian tubuh lainnya terutama di bagian lipatan seperti ketiak, kaki dll. Penyakit ini biasanya pada tahap awal berupa tumor dan bila tidak cepat diatasi akan berkembang ke stadium lanjut berupa kanker.. dan kanker juga berkembang cepat ke stadium yang makin lama makin tinggi.

Limfoma ini sebenarnya bisa segera diatasi, terutama bila masih dalam stadium awal. Pemicu utama dari Limfoma ini banyak faktor mulai dari makanan dan lingkungan. Makanan yang mengandung pemanis-pewarna-pengawet buatan, vetsin, softdrink,makanan yang dibakar, ikan asin, daging dll. sedangkan lingkungan bisa menjadi pemicu kalau lingkungan tersebut terkontaminasi oleh logam berat.

Limfoma tidak saja diidap oleh orang dewasa tetapi juga bisa diidap oleh anak-anak, hal ini karena dipengaruhi oleh jenis makanan yang dimakan mengandung zat-zat yang menjadi pemicu terjadinya limfoma.

Pengobatn secara medis biasanya dengan melakuan penyinaran atau paling akhir dengan kemoterapi. bagi masyarakat yang biasa ke-2 pengobatan ini akan sangat membebani karena biaya yang sangat besar dan hasilnya kadang belum pasti.

Berdasarkan pengalaman saya selama ini, saya sering membantu masyarakat yang mencari alternatif herbal untuk menyembuhkan atau meringankan sakit akibat limfoma tersebut. Herbal yang saya rekomendasikan sudah standar yaitu :
1. Lhiforcan (keladi tikus, temu putih dan sambiloto)
2. Tapak dara
3. Rumput mutiara
Semua sdh bentuk kapsul sehingga memudahkan untuk dikonsumsi.Semua herbal tsb diminum selama 1 bulan penuh dgn dosis 3x1-2 (tergantung dari kondisi sakitnya), kemudian jeda 1 minggu dan lanjutkan lagi.

Dari beberapa informasi orang yang konsumsi herbal tersebut, ternyata hasilnya sangat memuaskan dan alhamdulillah sudah banyak yang teratasi. Berikut informasi yang disampaikan via tlp k saya sbb :
Pagi ini, jam 10.00 WIB tgl 21 Juli 2010 sy ditelepon Bp. Wasidi- Sragen,Jawa Tengah (085711284440) salah seorang pemesan herbal utk LIMFOMA, beliau ingin pesan lagi herbal utk limfoma. setelah sy tanyakan kondisi terakhirnya, beliau menyatakan banyak kemajuan setelah minum herbal tsb


Wass
Prakoso,budi
081310343598
YM : budi_prakoso98

Menghancurkan Batu Ginjal tanpa Operasi?

Batu ginjal merupakan salah satu sakit yang sangat berat bagi penderitanya karena penderita akan merasakan sakit yang luar biasa pada saat akan berkemih. Batu ginjal ini merupakan hasil endapan dari residu yang mengumpal sehingga membentuk batuan yang menghambat disaluran ginjal maupun di ginjalnya sendiri.

Satu-satunya solusi yang bisa meredakan rasa sakit tersebut adalah dengan penghancuran batu ginjal tersebut. Ada beberapa alternatif untuk menghancurkan batu ginjal tersebut :
1. Operasi mengangkatan batu ginjal
2. Penembakan dgn sinar
3. Alternatif dengan herbal

Untuk point 1 & 2 sudah umum dilakukan oleh kalangan medis, untuk menghancurkan batu ginjal tersebut biasanya menghabiskan biaya yang besar sehingga sangat memberatkan bagi masyarakat yang tidak mempunyai biaya.

Solusi terakhir yang memerlukan biaya sangat terjangkau namun sangat efektif untuk menghancurkan batu ginjal tersebut. Namun demikian untuk alternatif herbal ini juga harus hati-hati karena tidak semua herbal bisa digunakan dan harus paham karakteristik dari herbal itu sendiri. Herbal yang secara riset bisa menghancurkan batu ginjal adalah : daun sendok, kumis kucing, meniran, tempuyung, keji beling. Utk herbal terakhir sebaiknya jangan dikonsumsi karena efek dari penghancurannya menyebabkan batu ginjal hancur dalam kondisi besar dan tajam sehingga akan merobek saluran kemih, dan pada akhirnya akan melukai dan terjadi pendarahan.

Herbal yang sangat dianjurkan untuk dikonsumsi untk mengatasi batu ginjal adalah :
1. Kumis kucing
2. Tempuyung
3. Meniran

Berdasarkan pengalaman selama ini, saya selalu merekomendasikan herbal tersebut untk atasi batu ginjal. Hasilnya batu ginjal hancur dalam waktu paling singkat 1 (satu) hari sejak pertama kali minum dan beberapa hari/minggu, hal ini tergantung juga dari besarnya batu ginjal tersebut. Namun, dengan rekomendasi herbal tersebut di atas, Insyaallah batu ginjal dapat di atasi.

Wass,

Prakoso, budi
YM : budi_prakoso98
081310343598
budiprakoso98@gmail.com