How many calories are in hawker dishes? I try this AI tool to see if it’s accurate, Lifestyle News

How many calories are in hawker dishes? I try this AI tool to see if it’s accurate, Lifestyle News


Like many others, my New Year’s resolution is to start eating healthier — a big part of my plan involves tracking how many calories I’m consuming in a day.

But as I started logging my calorie intake with an AI calorie tracker, I realised that it may not be as accurate in tracking the nutritional value of local foods, as most trackers cater to Western markets.

Just as I was about to give up my favourite cai fan and char kway teow altogether, I came across HawkerSense — an AI calorie tracking tool that claims to accurately read local cuisine, even detecting details such as the amount of oil used in ban mian or curry gravy poured over rice.

Wong Qi Han, product lead at Google and the developer of HawkerSense, told AsiaOne the reason for developing the tool is to help Singaporeans solve the problem of not being able to accurately track the nutritional content in their hawker meals.

“When I looked at a plate of cai fan, I thought it was an interesting problem for computer vision. Captcha squares would fail at labelling a plate correctly too many overlapping items, ambiguous ingredients. So I took it as an engineering challenge,” he said.

The tool also helps Singaporeans to make healthier choices without having to give up their beloved hawker fare.

“I want to help people order smarter without giving up what they love. It’s about awareness, not restriction,” Qi Han added.

According to him, HawkerSense’s core prototype took two weeks to build and saw close to 3,000 meals scanned. 20 days after the tool was launched, it tracked a total of 1.8 million calories. 

In addition to identifying local dishes accurately, Qi Han also designed the tool to be more engaging for users — using the relatable “Nutri-Grade” labels and Singlish. 

“A clinical ‘high fat content’ warning is easy to ignore. A localised tip like, ‘Wah, a bit oily lah. Ask for less gravy next time!’ hits different. It turns the AI from a judge into a helpful buddy, making the advice more palatable and certainly more memorable than a standard nutrition label,” he explained.

Sensing hawker fare

Determined to stick to my New Year’s resolution, I decided to give the tool a go to see if it’s accurate. 

For my experiment, I compared the readings of three hawker meals from a regular AI calorie tracker and HawkerSense. 

Besides calories, both tools tracked nutritional information including protein, carbohydrates and fat, as well as the percentage each nutritional value takes up with reference to the daily recommended intake. 

The first meal is a local favourite — wonton mee with soup and a side of fried wontons.

After uploading a photo of the meal to HawkerSense, the tool not only calculated the calories in the meal, but also provided insights into “blindspots” such as how the noodles are likely cooked in lard, which adds significant hidden saturated fat.

It also correctly identified most of the meal, down to the bok choy and clear soup.

The entire meal came up to 950 kcals, labelled as “Nutri-Grade C” according to HawkerSense.

While most may recognise Nutri-Grade as Singapore’s labelling system that identifies levels of sugar (with sodium and saturated fats coming in 2027) in beverages, the tool uses it as a relatable frame of reference for users. 

When I uploaded the same picture to a generic AI calorie tracker, the meal registered as having a mere 592 kcals. 





Read Full Article At Source