Local Stanford Grad Made Chicken Rice Classifier In Less Than 4 Hours
Mention “chicken rice” for lunch and I bet my bottom dollar, you’ll be met with euphoric shouts of approval all around.
Singaporean’s love for chicken rice is unparalleled. We breathe and bleed chicken rice.
With this in mind, Preston Lim, an Associate Software Engineer at GovTech Singapore, decided to do up a machine classifier to help us differentiate between steamed and roasted chicken rice pictures.
Purely for laughs.
Inspired by kuey.png account
The entire frivolous episode started during a lunchtime conversation with brought up a hilarious Instagram account called kuey.png.
The owner of the account uploads a different picture of chicken rice everyday – almost religiously, garnering an impressive feed and followers count.
Somebody from the group then made a casual remark,
Wow this would be an amazing machine learning dataset for chicken rice!
That’s all it took for the cogs in Mr Lim’s brain to start whirring.
Preston decided to create a machine learning classifier that allows you to take or upload a photo of chicken rice.
The programme would determine and recognise if your picture was a plate of steamed or roasted chicken rice.
Programme could recognise roast chicken pics in 4 hours
For readers who want a comprehensive explanation of how the machine learning classifier got invented, the tl;dr version is right here.
Here’s a quick rundown of what he did.
1. Painstakingly screenshot 279 individual pictures of chicken rice from @kuey.png on Instagram, with permission of course.
2. Uploaded the images onto AutoML Vision and labelled them as either roasted or steamed.
3. Let AutoML train the model while he binged on Netflix. Apparently, the software did several labour intensive tasks automatically, which reduced his curation time.
4. The chicken rice classifier produced its first round of results after 30 minutes.
The machine was able to identify all the roast chicken rice images accurately (yay!).
But it got a tad confused when it comes to steamed chicken rice, especially those that were slathered with too much chilli or soy sauce.
Well, upon closer inspection, we can see why some of the photos would be bewildering to a programme.
Are you eating chicken rice with chilli or chilli with chicken rice?
Preston tested the model with new images of chicken rice and the machine was able to classify them precisely — though higher accuracy is yielded for roasted chicken rice.
Local version of hilarious ‘Not Hotdog’ app?
Preston’s efforts definitely reminded us of the “Not Hotdog” app that went viral recently.
The app’s sole purpose it to let you snap a picture of a random object, and allow the app to tell you whether it thinks that image is a hotdog or not.
Though the team dedicated unprecedented amounts of resources to recognising hotdogs, the app continues to fail in hilarious ways.
Steam or roast, we love them both
We didn’t even know we needed the chicken rice classifier in our lives till we heard of this.
So thank you Preston for bestowing this upon us.