-
In this project, I conducted first-hand and secondary research on Machine Learning (ML) and Artificial Intelligence (AI) technology to develop a deeper understanding of how algorithms and ML systems perceive us and our surroundings.
The problem with ML today is the lack of data available to train machines to recognize complex ideas an image can contain and to recognize how people will naturally interpret them. Training datasets should grow, change, and evolve the same way a picture can move from one person to another by sharing it and contributing thoughts to it as it passes along over time and space. The phrase "an image is worth a thousand words" is a limiting description because it neglects to mention an image's cultural significance or historical relevancy, which is worth more than (a thousand) words.
In ML, a [word] is represented as an [object] in an image. For example, in captcha systems, you are asked to identify the [traffic cones] in an image or a set of images. In this case, the image is defined, or rather boxed, into one word, the object seen. The quality of the data, representing the object, is as essential as the quantity of the data.
How can symbolism be represented? What if we replace [object] with [artifact] or [monument]? How does that [word] affect the meaning associated with that [object]? What are the surfaces and essences of that [word]? How can we represent what can't be seen but can be known in an image?
[JeepneyML] is an open-source tool for building, capturing, and collecting synthetic image data to train an image recognition model to recognize jeepneys from other vehicles on the road. The tool consists of a [Jeepney Builder], which allows you to customize your own jeepney vehicle and place it in a synthetic environment of your choosing, and a [Training Library], where you can contribute the image data you captured in the builder to a [Training Dataset] for jeepney recognition training.
Share your captures by uploading them into a specific image data collection for training an object recognition model about [ jeepney vehicles ].
Today, image recognition machines are not able to capture the unique visual language of [jeepney] vehicles. It's not a [truck] or a [bus]. It's a cultural [artifact] adorned with symbolism, identity, and history. An image of a [jeepney] is rich with visual information. They display a personality that an object can possess. We are building technology that cannot sense the cultural attributes that give objects in an image their [name].
Jeepney's history began at the end of World War II when it was first recognized as a military vehicle left behind by American soldiers after the war. It has gone through some—mildly dramatic—changes since then.
A jeepney was and is a 1943 Willys Jeep.
Over the years, Jeepneys have turned into a cultural symbol in the Philippines. Today, they are still the same old Willys Jeep but all dressed up with an extended body capable of hoarding 20+ passengers both inside and outside the vehicle. They are recognized for their vibrant colors and kitschy ornamentation that jeepney drivers use to stand out from the mass jungle and to tell their story. These narratives include references to family members working overseas, nature scenes of their province and heritage, basketball players, and images that emphasize the Filipino obsession with Western pop culture.
A jeepney is a symbol containing many symbols.
Despite the jeepneys long-standing popular appeal in economy and mobility, they bear a fair amount of responsibility for contributing to the traffic and pollution prevalent in Metro Manila—the most densely populated city in the Philippines. It is a [symbol/object] near its end of use. What if ML technology can be a function of recognizing in order to preserve a [thing] we hold dear, and can't seem to let go of. The importance of community is in the ability to share similar values we hold for an [object]. With JeepneyML people can engage in a communal effort to train machines to see how important objects are to the identity of a place or persons—what a jeepney is to us.
Maybe, then we can finally let jeepneys go.
