Sky segmentation

Sky segmentation algorithm, which is based on application of deep learning segmentation methods. The algorithm successfully processes photos with trees, buildings and small objects, different lighting (with daylight or night) and can be used on the mobile device.
Problem
Trends in the field of photo editing change in fast way. And every year new photo/video editing apps come to the market. Sky segmentation is one of the popular filters, which has big demand among mass auditory and it's popularity increases every year.
Solution

We used our knowledge in deep learning and computer vision to solve this problem. Particularly, we experimented with different Image segmentation approaches and implemented own solution based on U-Net with ResNet Encoder. We also applied post processing algorithm which increases quality of result.

For model training we manually labeled 5000 photos from iPhone, Android.