TagCow - Blog

Content Filtering

mdroz — Sat, 25 Oct 2008 22:53:37 +0000

We recently expanded our Enterprise Tagging Platform to do near real-time content filtering of images, blogs, comments and advertisements. To give you a clear understanding of how this service can be useful I’ve developed a simple use case:

Use Case 1: Allowing for custom display ads

Background: On October 12, 2008 MySpace launched the self-serve advertising platform myAds which allows advertisers to create and run display ads on MySpace’s website. The advertisers can choose between pre-fabricated display ad templates or submit their own custom templates and then target those ads to specific demographics on MySpace. You can read about it in detail on TechCrunch: myAds Overview

Problem: Many companies allow for pre-fabricated templates to be used as the basis for display ads, profile backgrounds and photo album themes, but how can they possibly give millions of users the option to submit their own templates without exposing other users and advertisers to mature or inappropriate content?

Solution: Integrating Tagcow’s content filtering service into the “user generated template” submission and approval process. Once the custom template is submitted it would be pushed to our platform where it could be filtered against our client’s acceptable content guidelines and then returned to the client’s system with a status of approved or rejected.

Potential Users: Online Newspaper Sites, User Generated Art Sites, Photo Sharing Sites, Social Networking & Dating Sites and Online Media Companies implementing user generated content driven marketing campaigns.

Token diagram:

Thank You Taggers!

mdroz — Sat, 25 Oct 2008 15:38:31 +0000

We continue to make significant progress on-boarding companies to our Enterprise Tagging Platform. Last night I was doing a quick quality check on a job that was submitted for a well known museum and was completely floored by the quality of work that is being submitted by our workers. This enterprise tagging job had some serious challenges; first it requires a minimum of 10 tags per image and second a lot of the images are of simple objects from the museum (some of which are not very exciting). I tried tagging some of the images myself and could barely do it and definitely couldn’t do a very good job of it. With that I was a little afraid that we wouldn’t be able to produce quality results cost effectively but boy was I wrong. Below is an example image and subsequent tags submitted by one of our workers. It’s not mind blowing, but it’s exactly what is called for (quality and cost effective).

Oh yeah…and the main point of this post is to send out a huge THANK YOU to all of our workers. Without you we’re nothing. You enable everything we do. Tens of thousands of human taggers enabling enterprise tagging of millions and millions of images! Sweet!

Tags: abstract, industrial art, nonrepresentational, metal, sculpture, angular, geometric sharp, surfaces, steel, rust, weld, welds, welding, welded, punch hole

I can’t help but include this tagging example from our consumer service which reiterates the quality tagging done by our workers.

Tags: Michael Droz, mjdchild, 1980s, hungry hippos, batman, bop bag, crayola crayons, mickey mouse, candy, santa, christmas, gifts

The benefits of image tagging

twright — Thu, 02 Oct 2008 01:28:15 +0000

It is commonly known that search engines are currently incapable to identifying objects or content within the image and although there has been much work recently for computers to perform this work, the technology is not quite there. This problem leads to one of the most common questions we see pertaining to the value of image or photo tagging. The answer to this question depends on the solution you are seeking. General users with personal image collections have different needs than those of enterprise customers but the problem (and solution) is the same.

Personal

Personal or power users typically have growing collections of images that are stored in a flat directory structure. Unless they are disciplined or exact about how those images are stored and cataloged, images are stored by date and maybe by event. However, most users don’t go so far as to change the image naming structure so even when they are organized by date, there is no detail on what is in the photo. What they end up with is a bunch of images with file names like “DSC001248.jpg.” Not very descriptive to say the least! What this means to the user is if they want to find all pictures of their daughter, they must rely on their memory or spend time looking through thumbnail images to find images they are searching for. While this may be a viable solution, it is not very practical as their collection grows from 100’s to 1000’s of digital images over time.

Enterprise

Enterprise customers have a similar problem except theirs can cost them money. Enterprises have large collections of images that are stored in repositories. Although they tend to do a better job at renaming their images with names that relate to the content in the image, it is not practical to list out every detail in the image and include that in the file name. Alternatively, some enterprise customers will create elaborate directory structures to account for the insufficiency in identifying image content. However, as the repositories grow, enterprise customers are faced with an ever growing collection of images that are essentially lost or undiscovered because of the shear number of images in the repository. This can cost a company sales especially if a customer is unable to find a particular piece of art, for example.

So what can they do. Tagcow was started to address this issue. Instead of waiting for the technology to arrive to discover the content within an image, we approached the problem from a human point of view. Humans are uniquely qualified to identify the content in an image and will, for the foreseeable future, be able to add tags, such as human emotions, that computers will find difficult to ever tag. That said, image tags will empower users, personal and enterprise, to search their images or photos in a way never before possible: by content.

Consumer Image Tagging User Survey

mdroz — Tue, 19 Aug 2008 23:55:21 +0000

Prior to changing our service over to a pay for play model we conducted a user survey of our free service and the results were very positive. They indicate that our users were generally excited, optimistic and satisfied with our service. Using these data points we confidently started charging for our service. Not that we had a choice because the operational cost of the free service were killing us. The thing is once we made the switch to a pay model our users’ perception of quality made a 180 degree turn. Based on the declining daily registration numbers and negative feedback we got from our user base it was obvious that quality is a function of perception. When users didn’t have to pay they really liked our quality when we asked them to shell out for it they had a very different opinion. The good news is that it’s clear that if we improve the quality of our service users will pay.

Below are some of the highlights of that user survey:

Customer API

mnichols — Tue, 05 Aug 2008 08:25:06 +0000

This has always been in the plans, and has been requested many times. Our customer API will allow developers to integrate the TagCow.com image and photo tagging service into their applications. We’ll also introduce reseller rates that will allow developers to purchase our service at a discounted rate. If you are interested in participating in the TagCow.com API Beta program please email mnichols@tagcow.com.

iPhone Application

mnichols — Tue, 05 Aug 2008 08:22:18 +0000

You didn’t think we’d be left out of the excitement around the next gen iPhone platform did you? Our application will allow users to associate their iPhone with their tagcow account allowing their iPhone photo gallery to be automatically tagged. The application will also allow users to search for photos using their tag cloud to quickly find and share photos. This is currently planned for release in August.

Worker API

mdroz — Mon, 14 Jul 2008 08:26:02 +0000

Our worker platform will augment our Mechanical Turk implementation by allowing 3rd party developers and worker networks to plug directly into our platform and request/process image and photo tagging work. There’s already 1 major web 2.0 site that has implemented our API and is looking to have their users tag our images as a way to earn credits on for their system…oh the implications!

Automatic Image and Photo Tagging

mdroz — Mon, 14 Jul 2008 08:19:15 +0000

Automatic Image and Photo Tagging

When we first launched our image and photo tagging service in late March of this year (2008) a lot of people speculated on how we had “solved” the image and photo tagging problem. Some thought we had developed a breakthrough in automatic image and photo tagging while others more knowledgeable about the problems facing automatic image and photo tagging chirped about how we must be using humans to do our tagging. Well the latter were correct; we were in fact leverage humans to drive a great percentage of our tags. What I found most interesting about this debate was that no one considered that we would employ both humans and the “best of breed” facial and object recognition software to solve the problem – which is exactly what we’re doing. So for those who weren’t paying attention here’s a quick glimpse into where we started, where we are now and where we are headed.

March 2008: 90% human tagging / 10% automatic image and photo tagging (knowledge abstraction technology)

May 2008: 80% human tagging / 20% automatic image and photo tagging (knowledge abstraction technology and inference tagging technology)

July 2008: Testing partnerships with the leading facial and object recognition researchers and corporations.

August 2008: Implementation of facial recognition software to augment our other tagging technologies. We anticipate that this will reduce human tagging of “people” to 50%.

September 2008: Implementation of object recognition software to augment our other tagging technologies. We anticipate that this will reduce human tagging of “objects” to 75%.

October 2008: 65% human tagging / 35% automatic image and photo tagging (knowledge abstraction technology, inference tagging technology, facial recognition software and object recognition software)

Our goal is to get to a 50/50 split by mid 2009. Further improvements in automatic image and photo tagging will be offset by advanced human tagging, so we anticipate the 50/50 split to continue on for the foreseeable future.

We started with humans because we aren’t in business to innovate – we’re in business to completely solve problems. Human taggers solve the image tagging problem – period! Now, you may not like it and you may not be willing to pay for it, but it does solve the problem. Once the problem was solved we were left with two easy business problems; how to reduce costs and how to improve quality. Reducing costs will be achieved through implementing proprietary and licensed automatic tagging software and improved quality will be achieved through the combination of proprietary software and advanced human tagging.

Michael Droz, Chief Business Architect, mdroz@tagcow.com, (888) 860-3024

Image tagging templates

mdroz — Sun, 13 Jul 2008 08:18:02 +0000

A note on image tagging templates…

Our enterprise image tagging service has taught us a lot about how we should extend our consumer service. When a large e-commerce company engages us in a significant (60,000 – 10 million) image tagging job the challenge is in understanding their taxonomy and how to abstract that into a template that can drive quality image tagging results. We’ve become really good at this and have recently realized that while each of our enterprise customer seems to have a different reason for tagging their images and a unique taxonomy the concept of “image tagging templates” extends to our consumer service. In fact we can improve our image tagging quality by leaps and bounds by creating a handful of templates that our consumer image tagging clients can use – templates such as vacation, birthday, Christmas, wedding, Halloween and bar mitzvah.

Once we’ve implement the standard “image tagging templates” the next leap in quality and flexibility will come in the form of a “image tagging template wizard” that will allow clients to create and share their own templates.

BTW – If you’re interested in our Enterprise Image Tagging solution please call or e-mail me. You need this service if you are currently employing your own taggers or if you’re working on an SEO campaign and are just starting to think about how you might tag thousands or millions of images. We can do your tagging job faster and cheaper than anyone else.

Michael Droz, Chief Business Architect, mdroz@tagcow.com, (888) 860-3024

Tag Provider API

mnichols — Thu, 10 Jul 2008 12:05:45 +0000

We are excited to announce the release of the TagCow.com Worker API Beta!

The TagCow.com Worker API allows networks of workers to retrieve work items and submit the results to the TagCow.com system.

If you have a network of workers that you think can excel at tagging photos and providing professional meta data services for our enterprize customers, please email businessdevelopment@tagcow.com to setup a trial.