--- Log opened Sun Oct 23 00:00:34 2011 | ||
-!- blackburn [~blackburn@188.122.252.220] has joined #shogun | 11:32 | |
blackburn | sonney2k: how it is going? :) | 17:05 |
---|---|---|
CIA-3 | shogun: Sergey Lisitsyn * re74284b / (16 files in 5 dirs): Improved arpack wrapper, introduced LLTSA - http://git.io/4v9b7Q | 17:28 |
CIA-3 | shogun: Sergey Lisitsyn * r1ca6b60 / (3 files): Added references to LLTSA and NPE, updated NPE impl - http://git.io/h7eGGg | 17:28 |
CIA-3 | shogun: Sergey Lisitsyn * r0358e2e / data : Updated data - http://git.io/V2hNMg | 17:56 |
CIA-3 | shogun: Sergey Lisitsyn * rf236e25 / applications/faces/faces_embedding.py : Added faces dimreduction and clustering application - http://git.io/zZtOqA | 19:23 |
CIA-3 | shogun: Sergey Lisitsyn * re7b34f1 / src/shogun/converter/DiffusionMaps.cpp : Removed normalization from DiffusionMaps - http://git.io/lC4Q7A | 19:23 |
CIA-3 | shogun: Sergey Lisitsyn * rbf0bf48 / (2 files): Added warning of LLTSA and NPE are hardly applicable to very high-dimensional data - http://git.io/GJrTpQ | 19:35 |
@sonney2k | blackburn, fine | 20:32 |
blackburn | sonney2k: I did a lot of things today :) | 20:32 |
@sonney2k | the orange people will implemt the python buffer protocol | 20:32 |
blackburn | nice | 20:32 |
@sonney2k | and maybe will add some wrappers such that all of shogun's machine's can be used in their framework | 20:33 |
@sonney2k | the other way round is difficult (their code is python...) | 20:33 |
@sonney2k | so the buffer protocol is what will help us to e.g. use their data loading etc routines | 20:33 |
@sonney2k | blackburn, then I talked to some llvm/clang guy | 20:34 |
blackburn | about? | 20:34 |
@sonney2k | and we discussed how to reduce memory requirements in compilation | 20:34 |
@sonney2k | clang requires only 1/2 of what gcc needed btw | 20:34 |
blackburn | hmm | 20:34 |
@sonney2k | so in the end I now have a plan and will try to split up the file | 20:35 |
CIA-3 | shogun: Soeren Sonnenburg * r89c91c4 / (src/shogun/io/InputParser.h src/shogun/io/SimpleFile.h): fix compiler errors occurring with clang - http://git.io/JPH_Fg | 20:35 |
blackburn | nice | 20:35 |
@sonney2k | blackburn, I also talked to the opencv guy (gary bradski) | 20:36 |
@sonney2k | they have some new input features they call 'orb' | 20:36 |
@sonney2k | so it makes sense to add some example based on these (or the more general concept) | 20:37 |
blackburn | sonney2k: orb? what is it? | 20:37 |
@sonney2k | blackburn, then we long discussions with all the science guys | 20:37 |
@sonney2k | about basically everything | 20:37 |
@sonney2k | was great | 20:37 |
@sonney2k | blackburn, sth like sift | 20:37 |
blackburn | I see :) | 20:37 |
blackburn | sonney2k: have you kicked chris? | 20:37 |
blackburn | :D | 20:37 |
@sonney2k | in what respect | 20:38 |
blackburn | in ass | 20:38 |
blackburn | :D | 20:38 |
@sonney2k | I told him that you want to talk to him | 20:38 |
@sonney2k | he said no time :) | 20:38 |
@sonney2k | over | 20:38 |
@sonney2k | well he said in november again | 20:38 |
@sonney2k | he will fly to NY from here | 20:39 |
@sonney2k | and so is pretty busy... | 20:39 |
blackburn | yeah I know | 20:39 |
blackburn | okay will try to finish paper by myself :D | 20:40 |
blackburn | sonney2k: how soon will you fly to berlin again? | 20:42 |
@sonney2k | tomorrow | 20:42 |
blackburn | sonney2k: I have integrated pics: http://dl.dropbox.com/u/10139213/shogun/faceswclustering.png | 20:44 |
blackburn | sonney2k: you should like it :D | 20:49 |
@sonney2k | network is slow :( | 20:52 |
-!- in3xes [~in3xes@180.149.49.227] has joined #shogun | 20:55 | |
blackburn | sonney2k: it is embedding by diffusion maps and clustering by kmeans | 21:02 |
@sonney2k | I have the figure now | 21:06 |
@sonney2k | the example is not perfect but mostly works | 21:06 |
@sonney2k | I mean she is smiling in the upper left | 21:06 |
@sonney2k | but this should be on the right side | 21:06 |
@sonney2k | maybe you can make it more simple ? | 21:06 |
@sonney2k | just same expression in face and then turn head to the left ..middle...right? | 21:06 |
blackburn | sonney2k: it depends on method applied | 21:08 |
blackburn | sonney2k: I have no time to take more photos more precisely :) | 21:08 |
blackburn | sonney2k: treat it as demo | 21:08 |
@sonney2k | but certainly nice :) | 21:09 |
blackburn | sonney2k: two new methods available: LLTSA and NPE | 21:14 |
blackburn | unfortunately they involves crazy complex task | 21:14 |
blackburn | e.g. for pics 200x200 it requires to solve 40000x40000 *generalized* eigenproblem | 21:14 |
-!- in3xes_ [~in3xes@180.149.49.227] has joined #shogun | 21:27 | |
-!- in3xes [~in3xes@180.149.49.227] has quit [Ping timeout: 240 seconds] | 21:30 | |
-!- in3xes__ [~in3xes@180.149.49.227] has joined #shogun | 21:44 | |
-!- in3xes_ [~in3xes@180.149.49.227] has quit [Ping timeout: 260 seconds] | 21:47 | |
-!- in3xes__ [~in3xes@180.149.49.227] has quit [Quit: Leaving] | 21:57 | |
blackburn | sonney2k: still around? | 23:14 |
@sonney2k | yes | 23:14 |
blackburn | sonney2k: I would need kernels between images | 23:14 |
blackburn | essentially I would need to have kernels between say 20x20 and 40x30 images | 23:15 |
blackburn | I hope to come up with any idea of that kernel | 23:15 |
blackburn | but | 23:15 |
blackburn | we need ImageFeatures or so | 23:15 |
blackburn | right? | 23:15 |
@sonney2k | no you could use gaussian kernel on bytefeatures | 23:17 |
blackburn | sonney2k: eh?? how to store images of different sizes in bytefeatures? | 23:17 |
@sonney2k | why not scale them to some standard size alwyas? | 23:18 |
blackburn | sonney2k: do you think it is always possible? | 23:18 |
@sonney2k | why not? | 23:18 |
blackburn | sonney2k: it leads to losing information | 23:19 |
blackburn | e.g. if we have 200x150 pic | 23:19 |
blackburn | and 20x15 | 23:19 |
blackburn | if we resize 200->20 we will lose too much | 23:19 |
blackburn | but I'm pretty sure there is a way to define kernel between it | 23:20 |
@sonney2k | then scale them all to some reasonable size - say 50x50 or so | 23:20 |
blackburn | sonney2k: so do you think it is the only way? | 23:22 |
blackburn | sonney2k: are you against image features? | 23:36 |
@sonney2k | blackburn, you mean 2d features? | 23:36 |
@sonney2k | that have arbitrary sizes? | 23:36 |
blackburn | sonney2k: yeah generally yes | 23:36 |
@sonney2k | like stringfeatures? | 23:36 |
blackburn | yes | 23:36 |
@sonney2k | well it will just explode the wrapped code even more :) | 23:37 |
blackburn | ehhh how? | 23:37 |
@sonney2k | another feature class for arbitrary types... | 23:37 |
@sonney2k | I got to reboot | 23:37 |
@sonney2k | brb | 23:37 |
blackburn | sonney2k: ok | 23:37 |
blackburn | sonney2k: I'm pretty sure it would have ~4 types | 23:50 |
blackburn | I don't think anybody wants to have float64_t 2d features | 23:50 |
@sonney2k | blackburn, well I don't mind | 23:54 |
blackburn | sonney2k: yokay | 23:54 |
blackburn | okay* | 23:54 |
blackburn | :D | 23:54 |
blackburn | sonney2k: I don't want to resize images because I've got to make some classifying on roadsigns dataset | 23:55 |
blackburn | and pictures are pretty bad | 23:56 |
blackburn | resizing could make it even worse | 23:56 |
blackburn | sonney2k: is it true "ich bin ein berliner" is a right german? | 23:59 |
blackburn | :) | 23:59 |
--- Log closed Mon Oct 24 00:00:40 2011 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!