introducing CoDiFi

iPad like interface

Relate anything through roles, events and properties

a search for bead brings up bead photos, specialists, texts

a search for burials

Playing with Faces in Aperture for LHOTH

I am working on reprocessing all of the photos and other media for the Last House on the Hill project with the help of Apples Aperture 3. The jury is still out on whether we can use the app. With 40,000 pictures in the library, the thumbnails, previews and database weighs in at 22GB. However, there are some great benefits to using the app for media management, and some ‘neat’ features, like Faces, which may prove more than novel and actually useful!

Faces in Aperture

By letting my computer grind over the 41,243 pictures in the media database overnight, Aperture produced previews and thumbnails of all media, whether tif, jpeg, raw, movie, very convenient. But it also analyzed the pics for facial features. You begin to build the lineup by clicking on any picture where Aperture has detected a face, and giving it a name.

Confirming Pedja

Once you have added a person, you can click on his/her face and then confirm other pictures of that person. Here I’m going through pix of Pedja. As you confirm or reject pix, the choices improve.

Actual photos including Pedja

The usefulness comes when you switch to the actual photos including the person. We can see a life history of the person in all of their contexts, and since this information is semi-automatically derived, we’re able to build out a person-centric view of the images that would otherwise take hours (likely days) to do.

As I’m going through the pix, Aperture makes it easy to add missing faces as I come across them. Just click on the pic and hit the N key, the dialog comes up and Aperture attempts to find faces in the shot. You can easily add more facial regions, as I’ve done in this example.

Adding missing faces

If a person is in your address book, their details are automagically linked, very handy. You can limit suggestions to a specific project. This is critical if your Aperture library contains many different projects – personal, weddings, trips, work – you can imagine what a mess of faces this would cause.

Kitchen area on Bastille Day, 2001

Archaeological Faces: Of course, it would be fabulous if we could use Aperture to recognize archaeological features automagically. Imagine being able to point Aperture at a set of pictures and have it analyze them for features like fire installations, obsidian caches and the like. Maybe someday.

For now, Aperture is a great tool for humanizing the massive image collection with relative ease. It’ll be another few sessions before I’ll know if the data we get out of Aperture is worth the time it takes to identify Faces (and soon, Places), but so far, it’s looking good.

LHOTH Processing Pipeline

Let us get back on the blog track with a slew of new updates. Today, we’ll cover our processing pipeline for moving data in and through the LHOTH database.

Here’s a video of the process in action:

Processing Pipeline in DropBox

It’s taken us a while to get this going, but the process works very well. We use DropBox for handling all of our files. If you don’t, you should check it out, it’s an amazing tool for collaboration, syncing your life and backing up your stuff.

Thanks to DropBox, any file in the pipeline can be viewed from the web, iPhone, iPad, or laptop, and is always in sync.

LHOTH on the iPhone

The Pipeline

1) Unprocessed: Any Source file that we want to add to the database goes here. It’s a staging area. Any team member can easily get to the file and see it, open it, check it out.

2) For Review: Team member figures out who needs to provide input on the source document and contacts the person for their review. For example, I (Michael) may want Ruth to review a document to make sure it’s the right version, or to agree on which fields we want to map. Nico Tripcevich may be asked to review spatial data for the purposes of linking maps to the database. After review, the database team will process the source doc to prepare it for integration in the database, and move this file to the In Process folder. If changes need to be made to the original (if DATA is changed, that is), then the original is copied to the Archive folder.

3) In Process: Actively worked on source docs. These docs are linked to the database, using Filemaker 11′s new recurring import feature.

4) Ready for Grinding: Once a source file is completely processed, it’s ready to be transformed into Event data as RDF. We fondly call this meat grinding, as in making sausage from the data. Our mapping process produces open data that can be output in a variety of formats. We use XML, but we can output reports, Excel spreadsheets, or prepare the data to be visualized in a relational database.

5) Archive: Process sources are archived, along with their original companions and the mapping instructions, assuring full empirical provenance.

Last Note on DropBox: We love the fact that DropBox automagically versions all documents, including the database, so if we ever make a mistake (ok, when we make mistakes), we can roll back files to previous versions. Sweet!

DropBox versioning of a file in the processing pipeline

Adding stuff the the MediaHub

Adding stuff the the MediaHub

 

UC Berkeley MediaHUB: Adding or Updating Files from Michael Ashley on Vimeo.

Excavating 1999

Today, I rolled up my sleeves (ok, it’s a metaphor, I am in shorts and a tshirt) and dug deep into the 1999 media and their metadata. A special shoutout to the original ‘metador’ Scott Calhoun, who worked (slaved) so hard to get the media mediated. Excellent work!

I am attempting to hand knit data from some 3000 images or so taken in 1999. Like most metadata projects of old, there’s a lot of munging that needs to get done to make this great resource of cool images useful to anyone. Making progress, though. Here’s a couple things I figured out in my manual natural language processing escapade.

Mavili and onlooking workers, photo by Casey Gordon

About 64% of the 1999 images have captions. There are 4483 images, and I’ve been able to revive metadata for 2851 of them so far. This is not to say all is lost. A lot of images are people, parties, a couple trips here and there. Plus, we shot find (artifacts) by including the label and may not have ‘slated’ the image in the log. Well, this at least is my working theory.

This all said, I am going to use 1999 as a base model for the semantic linking we want to achieve between people, things, and media in places. To do this, I am working out how to grind up the data into these lovely packages. I feel compelled to do this old school style, the algorithm is my brain, eyes and hands.

Here’s an example of the fun. In 1999, we used a field called context, which later became Area, such as the BACH area, or the Dig House. Doing some quick analysis, here’s the list of variations of contexts that needs to be cleaned up:

1999 Contexts

The variations will be nice to grind over for ‘Did you mean: Dig House?’ later, but for now, I’m working to rationalize the terms to a master list of places across the site and off. My favorites here are asterisk *, some offsite mudmaking, and EFES (the place, not the beer).

Mapping our minds

Welcome to the Last House on the Hill

 

Navigate this blog to understand the project, its premises, goals, and methods. It is all work in progress, so please feel free to contribute with your comment.

Highlights:

Premises and Goals 

Database Narratives

Ontology readings