As of 2018, Polarr Vision Engine includes 20+ core C.V. modules and neural network models that can be compounded into powerful consumers experiences such as guiding user to take better photos interactively on phones or converting arbitrary image from 2D to 3D in web browsers. Below are a few experiences we’re particularly proud of and we will keep adding more as we develop more experiences.

 

Polarr Machine Guided Composition

The demo shows Polarr Vision Engine's capability to automatically identify, track, remember and guide a user to take interesting frames of photos while the user scans a scene. All frames are automatically captured with filters automatically applied. The first half of the video is a user panning the phone by hand, the second half is the user holding the phone still toward a screen playing videos.

 

Polarr Object Removal in High-res Videos

The demo shows Polarr Vision Engine's capability to identify and erase single or multiple objects in videos with minimal obvious artifacts. The first segment shows two boats being erased from the ocean, where the second and the more challenging segments shows a person erased while crossing path with another person. The model currently runs on a MacBook Pro at 30fps, and is yet to be migrated to mobile devices.

 

Polarr Image 2D to 3D

The demo shows Polarr Vision Engine's capability to convert arbitrary 2D image to 3D inside web browser through a Polarr browser plugin. All computation happens offline on a 2017 MacBook Pro running Intel Core i5 CPU with the Intel embedded GPU.

 

Polarr Aesthetic Ranking

The demo shows Polarr Vision Engine's capability to detect similar images, then rank, and evaluate aesthetic qualities among groups of similar images inside Polarr Album+. Such capability enables Album+ to automatically favorite images, help people delete duplicates and offer a de-cluttered view of photo albums.

 

Polarr Color Grading through user-created QR codes

The demo shows Polarr Vision Engine's capability to consistently grade color effects through photos and videos. The demo first shows a user using Polarr Photo Editor to create a custom effect, and export the effect as a QR code. Then the user imports the QR code into Polarr's video app called 24fps. Polarr's QR code includes effects that can be re-used and shared across all applications carrying Polarr Vision Engine's rendering module.

As of November 2018, there are over 1 million QR filters created by Polarr Photo Editor's users and you can find these filters circulating on Instagram, Tumblr and Pinterest.