If you’ve ever faced the fundamentally onerous conundrum of figuring out if your chicken rice consists of roasted or steamed chicken, worry no further. A software engineer with a degree in bioengineering and a master’s degree in the field of computer science from Stanford University recently came up with a machine learning program that could a solution to Singapore’s most important predicament.
On his Medium page, GovTech associate software engineer Preston Lim wrote about how he was inspired by meme account A Different Picture of Chicken Rice Every Day. Which, in case the name isn’t clear enough, posts daily pictures of chicken rice on both Facebook and Instagram (appropriately called @kuey.png, because ‘kuey png’ is ‘chicken rice’ in Hokkien). The dude running the account keeps to his promise, we assure you — today is the 284th day of chicken rice snapshot in a row.
With over 280 images of chicken rice uploaded, Lim realized that @kuey.png actually provided the perfect machine learning dataset for the dish. He contacted the account’s admin for his blessing (“Sure please go ahead and use the images for greater good,” said @kuey.png), Lim went to work on a tool that could tell if a photo is a plate of steamed or roasted chicken rice. A similar plot involving AI food identifiers was actually on HBO’s tech-startup satire series Silicon Valley, in which a hot dog-identifying app is created (it’s either a hot dog or not a hot dog).
Anyway, Lim goes on to explain the tech that went behind creating his chicken rice identifier with terms that we don’t understand, but what you need to know is this — his project was a success. He trained the program with 279 painstakingly labeled photos of chicken rice and after a couple of hours of work, the tool was able to accurately classify new photos of the dish. Chicken rice with heaps of chili and soy sauce proved difficult for the machine to classify, however.
Read on about Lim’s fun little project and petition for the man to turn it into a publicly accessible web app to benefit mankind.
Reader Interactions