Sunday, 3 March 2019

ColouriseSG: AI-powered tool instantly colours old Singapore photos

Month-old free online software developed by GovTech team is a global hit, with 68,000 users
By Shabana Begum, The Straits Times, 2 Mar 2019

It all started when three specialists from national information technology agency GovTech wanted to bring history to life by adding splashes of colour to old Singapore photographs.

Throughout January, the men from GovTech's Data Science and Artificial Intelligence Division developed a deep learning model that colourises black-and-white photos.

Called, the free online tool, which went live in the first week of last month, takes about 31/2 seconds to add colour to an uploaded photograph. But the trio who created the AI-powered tool for an internal company hackathon did not realise that would gain international traction.

Just days after the tool was put up on the Internet, users from England to Argentina were raving about the tool's speed and colourising precision on Twitter.

"A Dutch user tweeted that he colourised the one photo he has of his mum when she was young," said Mr Tan Kai Wei, 26, one of's creators.

"There were a few who didn't know what their late grandparents looked like in reality. This tool gave a possibility for them to find out."

Mr Preston Lim, 26, the software engineer behind, said: "These anecdotal comments highlight how visceral and emotional colour is, and how much emotion colour can bring to people."

About 68,000 users globally have used the tool, with 270,000 colourised images generated.

Compared with existing colourisation tools, the month-old software is unique because it was intentionally trained to colour images specific to Singapore.

The deep learning model was fed with more than 500,000 colour photos from the 1980s onwards, from the National Archives of Singapore and the New York Public Library's Singapore collection. Eventually, the model could recognise Singapore-based objects and features to realistically shade a picture of a shophouse or a military procession.

"When we tested an old photo of a local schoolyard on a colourisation tool created in the United States, the colours were muted. On, the tones were more vibrant. The US-based tool was unable to recognise the elements or features within this Singaporean scene," said Mr Lim.

The model performs well on portraits because its training set comprised historical photos dominated by people, said co-creator Andrew Tan, 29, who worked on training the model.

The team is looking to provide the model as an open source to programmers who can run the tool, and they are hoping to collaborate with arts and heritage agencies to organise a colourisation photo exhibition.

The trio hope that will spur younger people to take an interest in the country's history.

"The National Archives has a wealth of old photos and a lot of people don't know about it. We thought that by colouring some of these photos, we would generate interest in looking back at some key moments in history," said Mr Andrew Tan.

While looking through digitised black-and-white photos, the team also encountered pictures of places that do not exist any more, for example, Jiksha Station, from an image of a trishaw going by rows of shophouses.

On top of evoking nostalgia and sparking an interest in history, can also act as a rough guide for digital colourists who may need help in choosing colours for certain objects.

Mr Ngoh Shian Bang, 24, a freelance photographer who has been dabbling in digital colourisation for a few months, said: " would save us a lot of time and we can focus on refining the colours that aren't accurate in a photo. It is particularly helpful if we were working on a busy street photo or a landscape picture with many elements."

Mr Lim said the tool should not replace traditional colourists. "It's best if a person uses machine learning to speed up their time or make their life easier. I think that's what the goal of artificial intelligence should be."

Mr Tan Kai Wei agreed: "It should be complementary rather than a substitute for the real human experience."

No comments:

Post a Comment