It would be great to have automatic categorisation of new uploads. -- John Vandenberg _______________________________________________ Commons-l mailing list Commons-l(a)lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/commons-l

とある白い猫, 20/09/2011 14:05: Yes, is it free software and is there a way to use Commons the other way round, ie as a tool to build (or improve) such a software, using the already categorised images? Nemo _______________________________________________ Commons-l mailing list Commons-l(a)lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/commons-l

** I've made some attempt to map images on Wikimedia commons to distinct concepts from DBpedia, see http://ookaboo.com/ This could be useful for forming a training set, but I haven't yet got around to releasing a public dump of the data. I have about 1 million things classified and could certainly extend the strategies used to get more. Unless there's been a really unprecedented breakthrough, I'd think that the application of machine vision to Wikimedia faces the problem of getting enough training data. If you had thousands or tens of thousands of photos that were labeled 'cat' or 'not cat', or 'member of plant species X' or 'not member of plant species X', you can train a classifier to make the distinction. However, if you've got two or three bad photos of a particular plant (which is what you have most of the times in Commons) you don't have enough training data to generalize. If you've got a specific mission, say genitals recognition, I think you can make progress, but to attack the general problem you need to go big with your training sets.

Every small category is a part of a big category. A system such as this will not be able to specify plant species, but it might well be able to find pictures of plants. If it then gives a list of plant pictures that are not in some plant category, animal pictures that are not in animal category, buildings that are not in a regional building category, maps that are not in a map of category, paintings that are not in a painter category, famous people that are not in a people category etcetera, it could deliver those to volunteers to further classify.

Certainly it is their first attempt in plant identification. They are dealing with very few species of plant as well. Such phone apps could eventually be very useful as for example you point your phone at... anything and it links to the wikipedia article - in the language setting of your phone! Consider this labs progress in the next few decades... To them we are just another image repository but unlike other image repositories we are seeking to tag images in less specialized manner. For instance an image repository for birds would generally avoid tagging birds by color but instead by species or some other scientific categorization, commons on the other hand wouldn't mind such categorization.

As I said before, people shouldn't expect perfect species identification overnight but perhaps this could be the result of research for the next few decades. They currently are considering dropping commons because they do not see any use for commons, we could show them potential fields of research using commons by basically telling them about our problems. For instance dealing with images we frequently delete (over copyright, trolling and etc) could easily be a task for them. -- とある白い猫 (To Aru Shiroi Neko) On Mon, Sep 26, 2011 at 20:39, Paul Houle <paul(a)ontology2.com> wrote: _______________________________________________ Commons-l mailing list Commons-l(a)lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/commons-l

Certainly it is their first attempt in plant identification. They are dealing with very few species of plant as well. Such phone apps could eventually be very useful as for example you point your phone at... anything and it links to the wikipedia article - in the language setting of your phone! Consider this labs progress in the next few decades... To them we are just another image repository but unlike other image repositories we are seeking to tag images in less specialized manner. For instance an image repository for birds would generally avoid tagging birds by color but instead by species or some other scientific categorization, commons on the other hand wouldn't mind such categorization. As I said before, people shouldn't expect perfect species identification overnight but perhaps this could be the result of research for the next few decades. They currently are considering dropping commons because they do not see any use for commons, we could show them potential fields of research using commons by basically telling them about our problems. For instance dealing with images we frequently delete (over copyright, trolling and etc) could easily be a task for them. -- とある白い猫 (To Aru Shiroi Neko)

** On 9/20/2011 9:36 AM, とある白い猫 wrote: The conference has no such tool yet, at least nothing we can use tomorrow but they are able to pretty accurately on what the images are. I am going to try to propose if they would be interested in providing commons with such a service, The website relevant is http://www.imageclef.org/2011/Wikipedia for commons but http://www.imageclef.org/2011/Plants is also interesting (even though it had nothing to do with commons so far. It could ver well be used for commons and wikispecies alike. -- とある白い猫 (To Aru Shiroi Neko) I've made some attempt to map images on Wikimedia commons to distinct concepts from DBpedia, see http://ookaboo.com/ This could be useful for forming a training set, but I haven't yet got around to releasing a public dump of the data. I have about 1 million things classified and could certainly extend the strategies used to get more. Unless there's been a really unprecedented breakthrough, I'd think that the application of machine vision to Wikimedia faces the problem of getting enough training data. If you had thousands or tens of thousands of photos that were labeled 'cat' or 'not cat', or 'member of plant species X' or 'not member of plant species X', you can train a classifier to make the distinction. However, if you've got two or three bad photos of a particular plant (which is what you have most of the times in Commons) you don't have enough training data to generalize. If you've got a specific mission, say genitals recognition, I think you can make progress, but to attack the general problem you need to go big with your training sets. _______________________________________________ Commons-l mailing list Commons-l(a)lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/commons-l