Fresh & New(er)

discussion of issues around digital media and museums by Seb Chan

Fresh & New(er) header image 2

Sharing with SepiaTown – historical images re-mapped

October 19th, 2010 by Seb Chan

Early in the year when I visited Josh Greenberg and the digital team at the New York Public Library, I was told about SepiaTown.

One of quite a few ‘Then & Now’ web projects (see also History Pin), SepiaTown puts historic images back on the (Google) map, also using Google Street View to connect the photography of yore with those of today.

We figured that we’d give SepiaTown a full collection of the geotagged images of Sydney from the sets we’d uploaded to our repository in the Commons on Flickr, and after a bit of to-ing and fro-ing we uploaded a datafile and waited.

We knew that quite a few of the geotags on these images were ‘approximations’, and that to properly do Then & Now, you also need to know the direction in which the photographer was facing. And we also knew that neither the metadata in Flickr or on our own system was enough.

So you can imagine our surprise when Jon Protas at SepiaTown popped into our inbox advising us –

We quickly discovered when we first dug into your collection’s geo-locations that many of them were mapped in a fairly general way, and fell short of our quality control levels. We spent the summer spot-checking each one and correcting the locations for almost all of the images you provided, and we are now confident that the vast majority of the images are now mapped within a 20 yard radius of the exact camera location (and are facing the right direction).

Some were quite tricky, but fortunately, site designer Eric Lehnartz, who is also our main uploader (and a bit of a geo-locating savant), was able to deduce even the more obscure and rural locations.

Wow. They’re improved, fixed, tweaked!

Not only that, SepiaTown are sending the corrected dataset back for ingestion into both our collection management system and thus also into Flickr.

Big thanks to Jon & Eric!

Here’s a few to try from the Powerhouse set (follow the link then click the Then/Now option):

Blaxland’s Tree
Sussex St, North from Market St
Erskine St, West from Kent St
The Spit, Middle Harbour

Check out their blog for other highlights. They have some fantastic images mapped in there.

Tags: 3 Comments

  • Hi Seb, that’s really neat (and incredibly good of them too!). Perhaps another reminder though of the need for a good geotagging tool for the general public to use so this can be crowd sourced rather than rely on a few keen and talented individuals?

    On another note, like History Pin this seems to rely on images being uploaded. Do you know of any sites that would allow you to point the service at a Flickr image (or other image hosting service) and automatically display the images and pick up any existing tagging from there, to at least take those steps out of the user workflow? And for that matter to use the Flickr API to send any amendments directly back?

    Just out of interest, I’m currently looking into automatic tagging and tag suggestions for a small personal project I’m working on (e.g. Zemanta, Yahoo, OpenCalais, Alchemy etc) and especially geotag suggestions based on text and tags on pages. It’s even possible using Yahoo Pipes, but signal to noise ratio is very poor! Do you know of anyone in the museums sector who has looked at this?

    Regards, James

    • Seb Chan

      Hi James

      The Flickr API should definitely be used by these projects to ingest the Commons images – it is probably more a lack of awareness than anything else.

      Text parsing -> automatic tag extraction -> geotagging – we tried this with OpenCalais two years ago and generated a considerable list of ‘places needing disambiguation’. The ones that were resolvable we used Geonames to locate. But now we will be using the Powerhouse API to finally connect these up to WOEIDs. As you point out, the big issue is noise – especially because parsing narrative text and extracting places in isolation removes them from the context which is what makes human disambiguation easy (and machine disambiguation hard). I know a couple of the projects on the weekend at Amped tried disambiguation to do provenance maps of our collection so there are a many eyes on the problem – and it is is only a matter of time.

      Still, the level of attention to detail that SepiaTown went to is going to be a human task for the forseeable future. Ask Museum of London about the hand-checking and aligning of places in Streetmuseum . . .

  • Thanks Seb. Looking at the different services, and especially WordPress extensions built around them, some do have the ability to extract terms in context, which should deliver some disambiguation. And perhaps crucially I did find one, I think it was Zamenda, that allowed you to set a variable to show (or at least detect in the API return) which of the terms terms were deemed to be geographic – if I post about my holiday it doesn’t mean I have been to http://www.geonames.org/maps/google_28.188_-82.74.html !