The chairman of Popsa, a photo printing start-up, Declan Mellett explained that his company aims to simplify the process of making photobooks. “It’s always been a pain to make photobooks because you had to have your images in certain places on your computer. We’re trying to make that process so simple. The first version of the app reduced the time it took from two hours to five minutes,” he said.
One of the strategic goals since the inception of the company was to integrate algorithms and machine learning. Mellett said: “From day one, it’s been the ambition to bring in a data science team. Otherwise, we would just be a photobook app. To make us different, we have to have machine learning and artificial intelligence, and data in the business. The trick is finding the right people.”
Enter Dr Chanuki Illushka Seresinhe. She has always loved working with neural networks and saw the job opportunity as the lead data scientist at Popsa as an ideal opportunity and perfect connection. Here, she can tackle classic computer vision tasks and use the experience of her PhD in which she was working a lot with neural networks and understanding beautiful places.
She said the focus at Popsa at the moment is the elimination of the tedious tasks involved in creating a photobook, such as finding and deleting the shots that are blurry, too dark, washed out or duplicates, as well as creating groupings by people.
Mellett said: “One of the thoughts we had is to eventually teach the app the image of your mother, therefore when it comes towards Mother’s Day, we can say ‘here’s a beautiful book for Mother’s Day’. It is really to take the process away from having to tap lots of images.”
Another thing Popsa does is create groupings of photos by location. They have to look more closely at behavioural data because many people turn off location services on their phones due to privacy concerns. “We have to be clever with how we find things,” she said.
But what about the human in the loop? How important is human intervention to the photobook curation process? “We suggest things. It’s always a suggestion. It’s not like we’ve definitely got it right. The way I’d like algorithms to work is where they are suggestive, so it is not that algorithms are taking over the work for you. It’s the humans and algorithms working together,” she said.
Although there are only two people in the data science team so far, Popsa has big plans to expand that over the coming year. Seresinhe explained that she was brought in to lay the data science foundation of the company, but driving forward the use of machine learning and algorithms will require cooperation from other parts of the 26-person business.
“We can’t do data science by ourselves. For example, we need the app engineers to build us testing prototypes and we need the platform engineers,” she said, explaining the importance of testing before and data science algorithm can go into a usable app.
“Facial detection is one of the very classic things that can go really, really wrong and definitely want to test that with a small test of users, to make sure everything is working. That’s a bit of a tedious process as a business but it is also very important one That’s the kind of thing that upsets me, the kind of thing I don’t want happening here,” she said.
She was referring to the inaccurate digital facial recognition which disproportionately affects people with darker skin tones. Mellet, who has many, many years’ experience in the photographic industry said this has been a problem back in the days of analogue photography. Essentially, white skin was the tone that colour film was developed to best represent.
The way to tackle this problem is to focus on it and train your algorithms to deal specifically with it, according to Seresinhe who is pleased to see that solving this is now a higher priority. “Obviously, it is a priority for me. It’s nice to see that suddenly the data science world is changing and they are starting to open their eyes to some of these issues.”
Alongside her data science experience, Seresinhe also has a background in design and training in behavioural economic science and is a visiting researcher at the Alan Turing Institute. I wondered how did working in research compared to working in a young enterprise.
“Things are faster which is more fun for me. I used to work before I did my PhD so I ran a digital design consultancy myself and I liked it because it was fast and I went to academia it was really frustrating; things can be very slow Everybody slows down into this very long term kind of research space."
She added: "I love working with tech and tech moves fast, especially artificial intelligence. There are all these new things coming along all the time so I like the startup world because there are always new challenges coming up, you get to play around with the cutting edge technology and that suits me.”