--- Log opened Tue Oct 02 00:00:17 2012 | ||
Guest88299 | Hello | 00:24 |
---|---|---|
Guest88299 | I'm trying to run shogun 2.0.0 with code and data that run fine under shogun 1.1.0 | 00:24 |
Guest88299 | but I get this error:TypeError: Cannot create new instances of type 'Labels' | 00:25 |
Guest88299 | I think this is the samep problem that is mentioned here:http://comments.gmane.org/gmane.comp.ai.machine-learning.shogun/2643 - but is unsolved | 00:25 |
Guest88299 | any ideas? | 00:25 |
shogun-buildbot | build #120 of nightly_default is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_default/builds/120 | 03:43 |
-!- Netsplit *.net <-> *.split quits: Guest88299 | 10:22 | |
-!- blackburn [5bdfb203@gateway/web/freenode/ip.91.223.178.3] has joined #shogun | 14:14 | |
wiking | blackburn: yo | 16:19 |
blackburn | wiking: hey | 16:19 |
wiking | blackburn: u were looking for me | 16:19 |
wiking | ? | 16:19 |
blackburn | yeah | 16:20 |
blackburn | these guys from company I am applying to want me to analyze https://imagecategorizer.services.open.xerox.com/ | 16:20 |
blackburn | I'd appreciate if you share any thoughts :) | 16:20 |
wiking | lol | 16:21 |
wiking | they've opensourced it? | 16:21 |
wiking | no | 16:21 |
wiking | it's just the service | 16:21 |
blackburn | heh right | 16:21 |
blackburn | that makes the task a task | 16:22 |
wiking | mmm | 16:22 |
wiking | it's quite nice | 16:22 |
wiking | have you tried to upload images? | 16:22 |
blackburn | yes, quite a lot already | 16:22 |
wiking | how is it? | 16:22 |
wiking | btw: it's fun to upload dummy images :D | 16:23 |
blackburn | well I put some blender images | 16:23 |
blackburn | and smoothed/inverted/colored images, etc | 16:23 |
wiking | i'm trying this http://412.laxallstars.com/files/2011/07/Brown.jpeg | 16:23 |
wiking | but it says there's high traffic now | 16:24 |
blackburn | I tried green plane | 16:24 |
wiking | and? | 16:24 |
blackburn | wanted it to think it is a grass | 16:24 |
blackburn | :D | 16:24 |
blackburn | no category at all | 16:24 |
wiking | mmm that's good | 16:24 |
blackburn | actually | 16:24 |
blackburn | check ILSVRC 2011 | 16:24 |
blackburn | these guys won it :) | 16:24 |
blackburn | probably it is the same system | 16:25 |
wiking | yeah i know their work | 16:25 |
wiking | mostly based on fisher vector | 16:25 |
wiking | right? | 16:25 |
blackburn | yes | 16:25 |
blackburn | product quantization | 16:25 |
wiking | yep | 16:25 |
blackburn | linear svm | 16:25 |
blackburn | and fisher vectors | 16:25 |
blackburn | SIFT + color histograms | 16:25 |
wiking | yep yep | 16:25 |
wiking | i know that one | 16:25 |
blackburn | do you know how can I prove that it is a sift? | 16:26 |
wiking | csurka is leading their group | 16:26 |
wiking | mmm good question | 16:26 |
blackburn | some computer vision hacking | 16:26 |
blackburn | I was thinking about it | 16:26 |
blackburn | but nothing except blurring | 16:26 |
wiking | hehehe | 16:26 |
wiking | well yeah u should trick sift | 16:26 |
wiking | well what could be actually a funny thing to do | 16:27 |
wiking | is to seed a random image | 16:27 |
wiking | with some patches/sift keypoints of a chicken picture | 16:27 |
wiking | :D | 16:27 |
blackburn | I want to backtrace feature building | 16:27 |
wiking | i bet that's possible | 16:27 |
blackburn | hmm | 16:27 |
blackburn | I thought of skating horse | 16:27 |
wiking | and that the image u construct is still not a chicken :) | 16:27 |
wiking | ehhehe | 16:27 |
blackburn | I will do that tomorrow | 16:27 |
blackburn | :D | 16:27 |
blackburn | well but still | 16:28 |
blackburn | I tried same thing actually | 16:28 |
blackburn | yesterday I took a hockey player image | 16:28 |
blackburn | and just 'cut' his legs | 16:28 |
blackburn | 'skating' category disappeared | 16:28 |
wiking | mmm | 16:29 |
wiking | niice | 16:29 |
blackburn | proves that it is a local feature thing | 16:29 |
wiking | so you think puting the skaters on a horse | 16:29 |
wiking | would do some trick :) | 16:29 |
wiking | as said | 16:29 |
blackburn | yeah horse on skates would make these guys laugh at least | 16:29 |
wiking | :) | 16:29 |
wiking | take a simple sift extractor | 16:29 |
wiking | and get some patches | 16:30 |
wiking | and randomly distribute that in a dummy image | 16:30 |
blackburn | why do you think it proves the SIFTness? | 16:30 |
wiking | and then maybe get 2 sets of differetn sift features (from 2 different image categories) and start randomly mixing them in a new dummy image | 16:31 |
blackburn | I am thinking about something very special about SIFT | 16:31 |
blackburn | blurring is one thing - gradients go worse with it | 16:31 |
wiking | well | 16:31 |
wiking | one good thing is an affine transform of the image | 16:32 |
wiking | to check what type of sift they are using | 16:32 |
wiking | the default sift implementation is not so good with affine transformation | 16:32 |
blackburn | hmm right right | 16:32 |
blackburn | rotations is what I didn't try yet | 16:32 |
blackburn | good idea | 16:32 |
wiking | well not only rotate | 16:33 |
blackburn | yeah | 16:33 |
wiking | but like rotate and make a trapezoid of the original image | 16:33 |
wiking | for example with the eiffel tower | 16:33 |
blackburn | well I'll try some fisheye | 16:33 |
wiking | yeah | 16:33 |
blackburn | to see whether it is robust | 16:33 |
blackburn | wiking: do you understand what is a way to handle multiple patches but within the same dimension of feature space? | 16:51 |
blackburn | that's the thing I miss | 16:51 |
wiking | blackburn: eh? | 16:53 |
wiking | blackburn: i guess they could use some pca | 16:53 |
wiking | as they usually do in their system at xerox | 16:53 |
blackburn | I mean if one image has 3 patches | 16:54 |
blackburn | and other has 9 | 16:54 |
blackburn | how pca could help there? | 16:54 |
wiking | well they don't give a damnfuck about the number of patches | 16:54 |
wiking | as they calculate in a way a frequency vector | 16:54 |
wiking | of the patches | 16:54 |
wiking | of the number of various patches | 16:55 |
blackburn | ahhh | 16:55 |
wiking | that is then normalised | 16:55 |
wiking | so that's your feature vector, i.e. the fisher vector | 16:55 |
wiking | in the case of xerox | 16:55 |
blackburn | so # of features is just ~ to # of possible visual words? | 16:55 |
wiking | yep yep | 16:55 |
wiking | usual drill | 16:55 |
blackburn | got it | 16:55 |
wiking | the fisher vector is something like a bag of visual words | 16:55 |
wiking | a bit more scientific | 16:55 |
wiking | but on the end it is basically doing that | 16:56 |
blackburn | cool | 16:56 |
blackburn | more clear now | 16:56 |
wiking | heheh welcome :) | 16:56 |
blackburn | thanks | 16:56 |
wiking | hack the shit out of their system :)) | 16:56 |
wiking | if i'd have time i'd really help | 16:56 |
blackburn | yeah, I should try to hack out sift | 16:56 |
wiking | because this way you could prove some funky security stuff | 16:56 |
blackburn | no problem you helped a lot already | 16:56 |
wiking | or let say | 16:56 |
wiking | how to SEO and image search engine :P | 16:57 |
blackburn | yeah | 16:57 |
wiking | afaik this has been not really exploited yet | 16:57 |
wiking | but for sure it can be | 16:57 |
wiking | as all the keypoint extractors | 16:57 |
wiking | are very very | 16:57 |
wiking | academic | 16:57 |
wiking | so they must have several flaws | 16:57 |
wiking | :) | 16:57 |
blackburn | wiking: what are you working on btw? | 17:06 |
-!- blackburn [5bdfb203@gateway/web/freenode/ip.91.223.178.3] has quit [Quit: Page closed] | 17:25 | |
-!- rebecca [a9e64c04@gateway/web/freenode/ip.169.230.76.4] has joined #shogun | 17:41 | |
-!- rebecca is now known as Guest44991 | 17:42 | |
-!- Guest44991 [a9e64c04@gateway/web/freenode/ip.169.230.76.4] has quit [Quit: Page closed] | 17:49 | |
-!- RebeccaTruty [a9e64c04@gateway/web/freenode/ip.169.230.76.4] has joined #shogun | 17:50 | |
RebeccaTruty | hey - anyone there that can answer a question? | 17:51 |
RebeccaTruty | this is "Guest88299" from overnight | 17:51 |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 18:02 | |
wiking | RebeccaTruty: hey | 18:55 |
wiking | RebeccaTruty: can u send the code maybe somewhere? like pastebin | 18:56 |
-!- blackburn [~blackburn@37.61.179.160] has joined #shogun | 19:17 | |
blackburn | wiking: https://dl.dropbox.com/u/10139213/ml/horses-snow_with_skates.jpg | 19:34 |
wiking | blackburn: hahhaha what are the tags? | 20:02 |
blackburn | wiking: skating, winter sports, donkey | 20:02 |
blackburn | donkey is the first for sure | 20:02 |
blackburn | but skating is here | 20:02 |
wiking | looool | 20:02 |
blackburn | feel myself a hacker | 20:02 |
blackburn | :D | 20:02 |
wiking | blackburn the hax0r | 20:03 |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Ping timeout: 246 seconds] | 20:05 | |
wiking | how's n4nd0's stuff with pthread? :) | 20:10 |
-!- blackburn [~blackburn@37.61.179.160] has quit [Quit: Leaving.] | 20:17 | |
-!- blackburn [~blackburn@37.61.179.160] has joined #shogun | 20:18 | |
RebeccaTruty | wiking: I put some code here | 20:53 |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 20:53 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: leaving] | 21:17 | |
wiking | RebeccaTruty: where? | 21:34 |
RebeccaTruty | oh sorry, here: http://pastebin.com/FczYqj40 | 21:39 |
wiking | RebeccaTruty: oh yeah | 21:39 |
wiking | the thing is that Labels class has been majorly redesign | 21:39 |
wiking | so you'll have to change your code | 21:39 |
RebeccaTruty | I see | 21:39 |
wiking | so in this case | 21:40 |
wiking | use DenseLabels | 21:40 |
wiking | wait | 21:40 |
wiking | random.choice([0.0, 1.0]) | 21:40 |
wiking | gives you 0 or 1 or a random number in [0,1] interval? | 21:40 |
RebeccaTruty | it give you a 1 or 0 | 21:40 |
wiking | ah ok | 21:41 |
wiking | then just use | 21:41 |
wiking | BinaryLabels | 21:41 |
RebeccaTruty | great, thanks | 21:41 |
wiking | http://www.shogun-toolbox.org/doc/en/current/classshogun_1_1CBinaryLabels.html | 21:41 |
wiking | that's the doxygen for the class | 21:41 |
wiking | but yeah it shoudl work | 21:42 |
wiking | if you just replace | 21:42 |
wiking | test_labels = Labels(np.array(raw_labels, dtype=np.double)) | 21:42 |
wiking | to | 21:42 |
RebeccaTruty | yup, thanks, sorry I didn't realize it was such a stupid question - I shoudl have looked into it more | 21:42 |
wiking | test_labels = BinaryLabels(np.array(raw_labels, dtype=np.double)) | 21:42 |
wiking | RebeccaTruty: na no worries | 21:42 |
wiking | we just needed a bit more complex labeling system... :P | 21:42 |
RebeccaTruty | understood. thanks for the help! | 21:42 |
wiking | no worries | 21:43 |
wiking | but next time i think you'd get a faster answer on the mailinglist of shogun | 21:43 |
wiking | mmm fuck where's the link on the new website to the mailing list :) | 21:44 |
wiking | sonney2k blackburn where's the mailing list information on the new website? | 21:44 |
wiking | ah ok i've found it | 21:45 |
wiking | http://www.shogun-toolbox.org/page/contact/others | 21:45 |
blackburn | uh you were having kinda discussion | 21:45 |
wiking | blackburn: ehhehe no worries forget that i was messaging :P | 21:45 |
blackburn | somebody asked about labels thing yesterday | 21:46 |
blackburn | was that you, RebeccaTruty | 21:46 |
blackburn | ? | 21:46 |
RebeccaTruty | yes - that was me | 21:46 |
blackburn | clear, I am sorry you've got to handle this transformation manually | 21:46 |
blackburn | we came to case where we had no other solution because of design gap | 21:47 |
RebeccaTruty | yeah, no problem, it should be an easy enough fix. | 21:48 |
wiking | blackburn: http://jrvis.com/trunk8/stars.html?rel=github | 22:12 |
blackburn | wiking: what's that/ | 22:13 |
wiking | stargazers :D | 22:13 |
blackburn | my city is not highlighted, is that good? :) | 22:14 |
blackburn | ahhh who stars his project | 22:14 |
blackburn | wiking: find me! | 22:15 |
blackburn | there are only 5 in russia | 22:15 |
blackburn | wiking: sad no one from north korea | 22:18 |
-!- RebeccaTruty [a9e64c04@gateway/web/freenode/ip.169.230.76.4] has left #shogun [] | 22:42 | |
wiking | http://en.wikipedia.org/wiki/Vasili_Blokhin#Mass_murder_of_Polish_officers | 22:42 |
wiking | this guy was crazy | 22:42 |
blackburn | wiking: heheh he has good chiefs | 22:46 |
blackburn | beria and yezhov | 22:46 |
blackburn | wiking: did you know that once beria was killed gov't sent an order to all libraries | 22:49 |
blackburn | to cut down beria article in encyclopedia | 22:49 |
blackburn | and replace it with beringov proliv | 22:49 |
wiking | lol | 22:55 |
-!- blackburn [~blackburn@37.61.179.160] has quit [Quit: Leaving.] | 23:01 | |
--- Log closed Wed Oct 03 00:00:17 2012 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!