Sift.pics (Category: Research, Creative / Artistic)

Name of Submitter(s): Vahur Puik, Lauri Elias
Name of Team: Ajapaik
Organisation: Estonian Photographic Heritage Society

Historic photographs in public collections very often lack basic classificatory tags – whether a picture is an interior or exterior view, has the picture been taken from ground level or from a raised viewpoint (a higher floor window or tower), is it a view photograph depicting mainly a place or a reportage photo about an event etc. Most of these decisions are easy to make for everyone only by looking at a picture, so it is useful to capture that information that would enable bulk filtering of images afterwards.
Sift.pics is a crowdsourcing application for that kind of binary categorisation about images. We take in pictures from public collections that users start to browse and ask them to make couple of sorting decisions, they can also favorite (bookmark) images, see the title of the image and open the image in an official web repository. The binary icon based categorisation is meant for maximum simplicity of the task (especially on a mobile device). Our experience shows that even children of the age of 4 can take interest in the tagging and make accurate decisions about the pictures.
Sift.pics is an infotaining app (web, android) that gives users a reason to go through pictures without looking for something specific but enabling unintentional discoveries. Apart from categorisation users favorites give an extra informational dimension to the images for the collection owners.
Crucial for the success of the app (as for any crowdsourcing project) is getting the users to use it, i.e. do the work. The more users have made all the 7 possible categorisations for each image the more trustworthy the results are. Currently we have more than 60000 decisions made for more than 5000 images (most tagged album has almost 27k decisions for 1129 pictures by 120 users).
At this stage Sift.pics has no artificial intelligence (for instance feature recognition) included, but the data gathered can be later used in combination with image recognition tools and used for training the algorithms. We believe the future combination of AI and crowdsourcing would make the app more efficient. Still the crowdsourcing component is central as it is also a way of communicating the collections, having people see the content.
URL for Entry: http://sift.pics, https://github.com/Ajapaik/sift-pics-web

Email: vahur@ajapaik.ee

Twitter: @SiftPics, @Ajapaik, @puik

Job Title: project manager, product owner

Background of Submitter:

We work as a full-time team since November 2014. Lauri is software developer doing the full stack for both Ajapaik and Sift.pics web apps (mobile development is bought in). Vahur has been working in several Estonian museums on posts related to exhibitions and photographic collections, he is the founding member of the Estonian Photographic Heritage Society. He is the visionary and product owner of both Ajapaik and Sift.pics. He presented the idea of Ajapaik already in 2009 in Vilnius, LIthuania at a conference (see his slides: http://www.slideshare.net/puik/a-useful-game-rephotographing-historic-views-from-public-collections ), but the idea only started to materialize in 2011 during a hackathon. More recently Ajapaik won a round at the Europeana Creative Open Innovation Challenges (Vahur’s presentation at the Europeana Creative Culture Jam conference: https://vimeo.com/album/3501099/video/134879694 ). Ajapaik was a finalist at the Estonian national Best e-service 2015 competition.

Problem / Challenge Space:

Sift.pics is a spinoff project for our main project Ajapaik.ee that is a crowdsourcing geotagging and rephotography platform. While looking for new imagery we noticed that very often pictures in collections do not have the basic categorisations that are relevant for geotagging (interior/exterior for instance) or accessibility for rephotography (pictures taken from the ground or from a higher vantage point), not to talk about other simple categorisations (if and how many people are there on the picture). Having this kind of information about image collections improves the searchability of images and also helps to better understand the collections (how many what kind of images is there).

Approach / Methodology:

The main idea behind Sift.pics is incremental crowdsourcing and distributed work in the same way as many successful crowdsourcing project (by Zooniverse for instance) have already demonstrated. Decisions made by many people is very usable information and such a workflow is more scalable than having the describing of images made by the stuff of the collections owners only. As mentioned earlier we believe that combining the data with feature recognition algorithms would be even more efficient but already our basic app has given new information about the imagery we have included so far. We also use gamification as a method for user engagement. Sift.pics currently has just a basic statistics of how many users have made how many decisions and the user sees his/her place in the contributors leaderboard.

Extent of showcasing BL Digital Content:

We’ve included the Early Photographically Illustrated Books collection of images (1468 pictures) from the British Library’s collection on Sift.pics that is different from the other albums currently features on Sift.pics as it is not by a single author-photographer. Other albums so far have been by one author resulting in a thematic mapping of one author’s oeuvre.

Impact of Project:

Sift.pics was submitted to the first Finnish Cultural Heritage hackathon competion Hack4.fi in spring 2015 and received 2nd prize (500 €). In May Sift.pics got a special prize at the Estonian Mobile Summit.
Sift.pics was also presented (ad hoc) during a workshop at the Europeana Creative Culture Jam in Vienna in July and in Vahur Puik’s presentation at SOIMA2015 (http://www.soima2015.org/ ) in Brussels where it got a lot of positive feedback. Sift.pics will also be presented at the Museum Computer Network annual conference in Minneaplis (in Nov 2015).
Currently Sift.pics features 7 albums: 5 with Estonian content, one with Finnish photographs and one of the Early Photographically Illustrated Books photographs from the BL. Although the needed categorisations are quite abstract and everybody can make decisions about unfamiliar photographs we tend to think that audience also has bigger emotional closeness to content they are more familiar with. As of September 14, 2015 we’ve had 187 individual user contributing categorisations.
In September 2015 Sift.pics received a grant of 5000€ from the Estonian Ministry of Culture.

Issues / Challenges faced during project(s):

Crowdsourcing initiatives depend on users and our main challenges are related to creating traction and user base for the applications (both Ajapaik.ee and Sift.pics). It is very much also connected to the user interface design issues and aspects of gamification that we use for both of the projects.
Also as our team is working under a non-profit organisation Estonian Photographic Heritage Society (no collection, no permanent stuff, just network of people working with photographic collections in Estonia) and both our projects Ajapaik.ee and Sift.pics have no revenue models in place, but have been developed with the help of public grants from Estonia one of the main challenges is sustainability.