Even if you’re not a machinist, you’ve probably had a crisis at home or with your car where you only needed one weird, tiny screw to fix the problem. You bring the part to a store and stand in line only to discover the part isn’t in stock. So they call it in, and the part that arrives maybe a week later is the wrong one. Then the whole process starts over again.
It all sounds terribly inefficient! So of course, there’s a new startup to help move the parts-finding world into the 21st century with a visual search tool for finding replacement parts. It’s called PartPic. Using its very own image-recognition software, PartPic works with suppliers to not only digitize their catalogs, but make all the parts inside searchable to customers using smartphones. With this new system, instead of trying to find a match in a phonebook-thick parts catalog, you just take a couple of pictures of the part with your smartphone and run a search. PartPic’s software will match your photo with the supplier’s photo to find the part you need.
Of course, it’s not totally perfect yet. The software is still in beta, and PartPic claims its algorithms are currently correct about 80 percent of the time with the first match. Even getting to that level of accuracy is hard work—it’s why the company has four PhDs on staff and has published two peer-reviewed papers about its computer vision technology.
Computer vision, or image recognition, as it’s commonly called, has been around in various iterations for decades. It’s used in cameras to detect if there’s a face to focus on. Or in more complex applications like DeepFace, the facial recognition system by Facebook that can identify an individual based on their picture. Computer vision is also used in software for the blind.
“The basic task in computer vision is to take in a single image, which is an array of pixel values,” says Yali Amit, computer science professor at the University of Chicago. “And depending on the task, computer vision detects certain objects in the image or recognizes one or more objects in an image.”
But for being such an obviously useful technology, it’s not as advanced as one might imagine. Some may remember Google Photo’s horrible mistake last year when the company misidentified black people in images as gorillas; they issued an apology for the offense, noting that image recognition still has a long way to go.
Yali Amit says our eyes are still more capable than our smartphone cameras.
“It’s very rare in the analysis of regular images that a machine can do better than a human,” says Amit. “Lighting can also be a huge problem,” Amit added. If a photo is taken in low light, it could muddle with the correctness of the identification process.
PartPic recognizes these challenges, and understands the barrier they present.
“We’re identifying the exact match,” says PartPic co-founder Jason Crain. “It’s important if you’re looking for a 2.5 inch screw with zinc finish and a grade of five and a thread size of eighteen, and you don’t need anything other than that, for us to be able to identify those specific details to really be a solution for this industry.”
Of course, if you’re searching for a missing part—and therefore have nothing to photograph—then PartPic is less useful. But even just by improving the searchability of parts databases, it would be taking the industry several steps forward. Fully streamlining this process with image recognition could result in massive savings in time and resources. Consider a small parts supply company tasked with sifting through huge inventories of stock to locate a single tiny component hundreds of times a day.
There are other image recognition companies out there that work on enterprise level software, like Slyce, an image search company that can do things like identify the brand of clothing someone is wearing to help with shopping. Still, PartPic (which was recognized by the White House’s Demo Day for new tech businesses last year), is really coming from a specialized knowledge base of industrial manufacturing and parts supply. And that puts it at a major advantage for a range of industries that wrestle with the same problems.