Realistic sky replacement is on the horizon for Autoenhance.ai

Realistic sky replacement is on the horizon for Autoenhance.ai

Realistic sky replacement is on the horizon for Autoenhance.ai

Read about our latest technical challenge towards developing a photo editing AI for the UK property marketing industry.

Sky replacement is an important feature of photo editing software. On property portals such as Rightmove and Zoopla, an image must show what a property looks like on the best days of summer to attract a potential buyer. Yet, given the regular gloomy weather in the UK, this simple request can be difficult for photographers to achieve. Using an automated photo editing Artificial Intelligence (AI) can overcome this.

At Autoenhance.ai, we’ve been developing our sky replacement algorithm to create high-quality blue skies available in an instant. However, we’ve recently experienced technical challenges towards creating a realistic horizon.

Grey sky in the original left-hand image; blue sky, but unrealistic because there is no horizon, in the middle image; final picture with a summer sky going to the horizon in the right-hand image.Grey sky in the original left-hand image; blue sky, but unrealistic because there is no horizon, in the middle image; final picture with a summer sky going to the horizon in the right-hand image.

Our initial sky replacement

As we believed that most external property images are taken from the same angle, we initially created a horizon for sky replacement that we could apply to all external images that run through our photo editing AI. We located the static horizon in the bottom section of the image. However, as the angle that images are taken from differs picture to picture, our static horizon was unsuccessful.

Grey skies in the left-hand image; unrealistic sky replacement in the right-hand image.Grey skies in the left-hand image; unrealistic sky replacement in the right-hand image.

How we solved the problem

To overcome this technical problem, we had to calculate where the horizon should begin based on where the sky starts in the image and predict at which point the horizon filters out. Our first algorithm worked well but took over 10 seconds to calculate. This is not ideal when the whole picture ought to be enhanced in under a second. By calculating the horizon on an image 1/10 of the scale, we received accurate results. This took processing time down to 0.05 seconds.

Grey skies in the left-hand image; our algorithm calculates the realistic horizon in under a second as illustrated in the right-hand image.Grey skies in the left-hand image; our algorithm calculates the realistic horizon in under a second as illustrated in the right-hand image.

Our algorithm calculates the horizon instantly and applies this effect during sky replacement, resulting in a realistic sky and high-quality enhanced image.

We hope you enjoyed our recent update. Follow our blog to keep up with our news.

Click here to register your interest in our service.

HomeAPIPricingBlogDashboard

© Copyright 2021, Autoenhance.ai. All Rights Reserved.