Join us now at the IRC channel.
| TechrightsSocial | @schestowitz@joindiaspora.com: Take the Ubuntu User Survey 2012 Now http://news.softpedia.com/news/Take-the-Ubuntu-User-Survey-2012-Now-257837.shtml #ubuntu #gnu #linux | Mar 11 00:23 |
|---|---|---|
| TechrightsSocial | -> Title: Take the Ubuntu User Survey 2012 Now - Softpedia .::. Size~: 46.44 KB | Mar 11 00:23 |
| schestowitz | "Yes, Canonical does have that problem - ignoring diaspora, identi.ca, gnewbook, etc. - all which have a large GNU/Linux user base. " | Mar 11 00:24 |
| XFaCE | schestowitz: gnewbook;? | Mar 11 00:38 |
| schestowitz | perhaps | Mar 11 00:50 |
| *oiaohm (~oiaohm@139.brs0109.brs.iprimus.net.au) has joined #boycottnovell-social | Mar 11 02:41 | |
| *oiaohm has quit (Changing host) | Mar 11 02:41 | |
| *oiaohm (~oiaohm@unaffiliated/oiaohm) has joined #boycottnovell-social | Mar 11 02:41 | |
| *oiaohm has quit (Quit: by all) | Mar 11 03:55 | |
| *oiaohm (~oiaohm@unaffiliated/oiaohm) has joined #boycottnovell-social | Mar 11 06:00 | |
| *oiaohm has quit (Quit: by all) | Mar 11 06:45 | |
| *oiaohm (~oiaohm@unaffiliated/oiaohm) has joined #boycottnovell-social | Mar 11 07:56 | |
| schestowitz | https://joindiaspora.com/posts/1402586 | Mar 11 09:05 |
| TechrightsSocial | @schestowitz@joindiaspora.com: Apple is Becoming a Patents Cult http://techrights.org/2012/03/10/the-cult-of-swpats/ #apple #swpats | Mar 11 09:05 |
| TechrightsSocial | -> Title: Apple is Becoming a Patents Cult | Techrights .::. Size~: 85.86 KB | Mar 11 09:05 |
| schestowitz | "Becoming?" | Mar 11 09:05 |
| schestowitz | "Kill it! Kill it with FIRE!" https://joindiaspora.com/posts/1399182 | Mar 11 09:05 |
| TechrightsSocial | Not a Diaspora post? | Mar 11 09:05 |
| schestowitz | http://identi.ca/notice/91741155 | Mar 11 09:07 |
| TechrightsSocial | @Mikael (inscius)'s status on Saturday, 10-Mar-12 09:00:05 UTC - Identi.ca: @schestowitz Thumbs up for Wikipedia :) | Mar 11 09:07 |
| *schestowitz has quit (Remote host closed the connection) | Mar 11 09:24 | |
| *schestowitz (~schestowi@host86-147-16-124.range86-147.btcentralplus.com) has joined #boycottnovell-social | Mar 11 09:25 | |
| -NickServ-schestowitz!~schestowi@host86-147-16-124.range86-147.btcentralplus.com has just authenticated as you (schestowitz) | Mar 11 09:25 | |
| *schestowitz has quit (Changing host) | Mar 11 09:25 | |
| *schestowitz (~schestowi@unaffiliated/schestowitz) has joined #boycottnovell-social | Mar 11 09:25 | |
| *pidgin_log has quit (Quit: Leaving.) | Mar 11 09:25 | |
| *pidgin_log (~roy@host86-147-16-124.range86-147.btcentralplus.com) has joined #boycottnovell-social | Mar 11 09:26 | |
| *scientes has quit (Read error: Connection reset by peer) | Mar 11 10:12 | |
| schestowitz | http://www.techdirt.com/blog/innovation/articles/20120307/13221318026/gutenberg-ebook-once-again-bible-is-forefront-publishing-technology.shtml http://www.theatlantic.com/technology/archive/2012/02/translating-the-bible-151-into-an-e-book-that-works-on-any-phone/252842/ | Mar 11 10:40 |
| TechrightsSocial | Title: The Gutenberg eBook: Once Again, The Bible Is At The Forefront Of Publishing Technology | Techdirt .::. Size~: 133.43 KB | Mar 11 10:40 |
| TechrightsSocial | Title: Translating the BibleInto an E-Book That Works on Any Phone - Alan Jacobs - Technology - The Atlantic .::. Size~: 112.64 KB | Mar 11 10:40 |
| schestowitz | Diaspora has finally been fixed such that continuous scrolling works again in Konqueror and Rekonq, my main Web browsers | Mar 11 10:54 |
| schestowitz | Let's see if they fix some other stuff too | Mar 11 10:55 |
| MinceR | http://www.reddit.com/tb/qr22d | Mar 11 11:31 |
| TechrightsSocial | Title: The DRM Monster via reddit.com .::. Size~: 1.63 KB | Mar 11 11:31 |
| schestowitz | " | Mar 11 11:32 |
| schestowitz | > Roy, | Mar 11 11:32 |
| schestowitz | > Interpreting the research so far, is your conclusion that the GMDS | Mar 11 11:32 |
| schestowitz | > FMM/Diffusion for 3D face recognition | Mar 11 11:32 |
| schestowitz | > lead to inferior than state of the art results? | Mar 11 11:32 |
| schestowitz | It's quite hard to provide a short answer to this question. Based on every literature surveys that I've read (I have looked quite exhaustively for these and read them entirely), interpretation in 3-D only (no assistance whatsoever from 2-D data) typically gives not much better than 90% verification rate, at best. There were some strands of work that use PCA for that, BKK use GMDS, but all works that claim very high | Mar 11 11:32 |
| schestowitz | verification rates deal with very small datasets -- ones that are usually not public, either. A classification problem of one-to-many where "many" is 10 or 50 people does not say much. One could just optimise an algorithm to latch onto pertinent characteristics of each person. In PCA, for example, separating the clouds to would be simpler. The tables summarising the results reported by different groups usually contain a | Mar 11 11:32 |
| schestowitz | column to signify sizes, or put differently, the quality of validation experiments. | Mar 11 11:32 |
| schestowitz | The exception to the above is the one single paper from Australia -- a paper with quite a few typos, very impressive results, the admission that they use thousands of images from just 3 people (private data), and outcomes that I am unable to reproduce. I am -- quite humbly -- very sceptical of it, having worked on the problem of 3-D-only facial recognition for a while and read papers on the subject too. Bear in mind that Mian' | Mar 11 11:32 |
| schestowitz | s student succeeded his supervisor's doctoral work, wherein Mian used ICP to get just about 70% in terms of recognition/verification, if I recall correctly. I had a discussion about this with a retired professor yesterday. He too is a sceptic. | Mar 11 11:32 |
| schestowitz | It is well established that 2-D recognition is a lot easier and all those famed algorithms for this purpose rely a great deal on it. The premise for NIST was, there is a great untapped potential in 3-D, but it was yet unexplored and unfulfilled. | Mar 11 11:32 |
| schestowitz | So returning to the original question, the main point is, what are state-of-the-art results? If we assume that since 2009 we're looking at 97% or better for 3-D only (FRGC), then we are well behind with GMDS/FMM/Diffusion. To get 32 out of 33 right _even when dealing with harder cases_ (borderline) is just very hard, even if I had to do it with my own eyes and my innate skills for telling people apart (the brain is well wired | Mar 11 11:32 |
| schestowitz | for it). 32 out of 33 is achievable for simpler cases. I tried this at the lab by presenting to Anastasia some examples as such, and she too was unable to make the classification based on just 3-D. Expressions complicate this further. | Mar 11 11:32 |
| schestowitz | " | Mar 11 11:32 |
| schestowitz | MinceR: Hehe | Mar 11 11:32 |
| schestowitz | "We love you captain patchy" | Mar 11 11:33 |
| schestowitz | "Last month I left 3 printed literature surveys on 3-D verification. I am coming to the lab again next week. Today I run more experiments." | Mar 11 11:33 |
| schestowitz | seems like I'm just moving back to a biomed project, at long last | Mar 11 11:34 |
| schestowitz | I've just gotten followed by this dude [NSFW] https://joindiaspora.com/people/72d042eac222d205 | Mar 11 11:43 |
| TechrightsSocial | Title: The page you were looking for doesn't exist (404) .::. Size~: 1.22 KB | Mar 11 11:43 |
| MinceR | http://www.reddit.com/tb/qr1av | Mar 11 11:47 |
| TechrightsSocial | Title: the best beginning to any book I've read via reddit.com .::. Size~: 1.66 KB | Mar 11 11:47 |
| schestowitz | At times of election, opposition to the government policies or Obama is wrongly perceived as GOP promotion. Helps gag Obama disappointees. | Mar 11 12:01 |
| schestowitz | http://www.youtube.com/watch?feature=player_embedded&v=yXyqCuoRvaE#! | Mar 11 12:02 |
| TechrightsSocial | Title: Breaking News: Obama Put on Notice - YouTube .::. Size~: 134.7 KB | Mar 11 12:02 |
| *oiaohm has quit (Quit: by all) | Mar 11 12:17 | |
| schestowitz | Today I got rather bored, so I decided to go and do some sports. I met Kieran for a table tennis match and I won very narrowly, he's a very good player. I won by one set and just two points. I then went to the spa for over an hour. | Mar 11 15:31 |
| schestowitz | qu1j0t3: http://twitter.com/directhex/statuses/178802155426295809 | Mar 11 16:00 |
| TechrightsSocial | @directhex: This sums up how I feel about my "antichrist" status on Boycott Novell: http://t.co/IOzGOk4Q | Mar 11 16:00 |
| TechrightsSocial | -> Title: I'm Honored - NYTimes.com .::. Size~: 54.94 KB | Mar 11 16:00 |
| schestowitz | http://identi.ca/notice/91771693 | Mar 11 16:38 |
| TechrightsSocial | @George Hostler (ghostler)'s status on Sunday, 11-Mar-12 16:29:08 UTC - Identi.ca: #Linux 9.31% http://ur1.ca/8m5oh this is more plausible than the M$ evangelism FUD by Wintrolls in c.o.l.advocacy that Linux totally failed. | Mar 11 16:38 |
| TechrightsSocial | -> Title: Stats for browsers and operating systems accessing sutor.com | Blog | Bob Sutor .::. Size~: 52.81 KB | Mar 11 16:38 |
| schestowitz | I've just booked train tickets for my trip to London next weekend next week (Christening of Amy). I am aiming to take my vacation from April 30th until the 10th of May, provided all is ripe for days off work. | Mar 11 17:46 |
| schestowitz | Hmm... subs needed... A couple of weeks ago a colleague kindly agreed to cover for me while I attend the baptism of my goddaughter; it turns out I won't be back early enough on Sunday (18th) either. The parents organise dinner at 1pm. I ned to see if itwould it be possible for someone -- perhaps the same coleague -- to cover that night too, | Mar 11 18:39 |
| schestowitz | " | Mar 11 19:33 |
| schestowitz | I have revisited this work which is similar to work I recently reviewed for IJCV (also work from Texas University). They are using wavelets to fuse together data and characterise similarity based on that. I spent an hour or so studying results they reported for FRGC, e.g. http://nguyendangbinh.org/Proceedings/BMVC/2006/papers/387.pdf | Mar 11 19:33 |
| TechrightsSocial | Not a web page! Aborting application/pdf type | Mar 11 19:33 |
| schestowitz | "On the extensive FRGC v2 database, Chang et al. [7, 6] examined the effects of facial expressions using two different 3D recognition algorithms. They reported a 92% rank-one recognition rate. Husken et al. [11] presented a multimodal approach that uses hierarchical graph matching (HGM). They extended their HGM approach from 2D to 3D but the reported 3D performance is lower than the 2D equivalent. Their fusion, however, | Mar 11 19:33 |
| schestowitz | offers competitive results, 96.8% verification rate at 0.001 false acceptance rate (FAR), compared to 86.9% for the 3D only. Maurer et al. [18] also presented a multimodal approach tested on the FRGC v2 database and reported a 87% verification rate at 0.01 FAR." | Mar 11 19:33 |
| schestowitz | A student of Bovik writes in his thesis (page 86) that "The 2-D+3-D face | Mar 11 19:33 |
| schestowitz | recognition literature is very limited, and a majority of the algorithms in this category use conventional methods, such as eigenfaces, to create independent 2-D and 3-D recognition systems and then combine the matching scores by a fusion strategy." | Mar 11 19:33 |
| schestowitz | He used a lot a great deal of Gabor filters, fused them with other classifiers. | Mar 11 19:33 |
| schestowitz | " | Mar 11 19:33 |
| schestowitz | -me | Mar 11 19:33 |
| schestowitz | " | Mar 11 22:34 |
| schestowitz | There is a fusion of methods therein, one that attains a higher performance by pooling together different classifiers. Kakadiaris reported on the performance of other groups and demonstrated a classifier that reaches 95% verification rates in the presence of expressions, markedly higher without them. The approach transforms the 3-D data onto a 2-D domain much like the work on Canonical Forms by BBK, but on this very large | Mar 11 22:34 |
| schestowitz | dataset it showed high performance. | Mar 11 22:34 |
| schestowitz | I have been trying to augment the extent explored by GMDS and then identify the best fit. A lot of entropy goes to waste otherwise. The problem I am having today, the algorithm is not stable enough (keeps crashing) and even if eventually it works in a stable fashion, the theoretic framework meets hurdles at less than 95% in the presence of expressions, even if properly masked to mitigate the effect of expressions-induced | Mar 11 22:34 |
| schestowitz | deformation. I am doing the best I can given the tools I have at hand. If the 3-D-only tools (discriminants) are fused together, higher verification rates can be attained, but we don't worry about that for now. In fact, running the experiments on many thousands of pairs knowing that the performance would not be state-of-the-art seems a bit premature. | Mar 11 22:34 |
| schestowitz | " | Mar 11 22:34 |
| schestowitz | -,me | Mar 11 22:35 |
| MinceR | tl;dr | Mar 11 22:35 |
| schestowitz | :/ | Mar 11 22:38 |
| *abeNd-org has quit (Ping timeout: 260 seconds) | Mar 11 22:58 | |
Generated by irclog2html.py 2.6 by Marius Gedminas - find it at mg.pov.lt!