Google AI demonstrates the Expertise that can turn your Past Image to 3D Cinematic Photos

Google launched cinematic photos on December 2020 to help people remember some of their most precious moments. Cinematic photos can transform a single 2D photo that was taken in the past “into a more immersive 3D animation.”

The conversion of 2D to 3D images involves simulating camera movement and parallax by inferring 3D depictions in an image. Just like the past computational photography, cinematic photos require a depth map for the provision of information regarding the 3D structure of a scene.

The technology that sustains depth estimation on smartphone relies on a geometry method and a multi-view stereo to concurrently capture multiple images at different viewpoints in order to solve for the depth of objects.

Google AI trained a convolutional neural network to activate cinematic photos on pictures that were not captured in multi-view stereo. The process used an encoder-decoder architecture to speculate a depth map drawn from “a single RGB image.”

Median filtering was applied to enhance the “depth map’s accuracy” and infer segmentation masks of any people in the photo using a DeepLab segmentation prototype trained on the Open Images dataset. 

According to a blog post published by Google, cinematic photos came into effect after collaboration between Google Photos teams and Google Research.