SuperSlowMo

Python

ML pipeline to create super slow motion using frame interpolation


SuperSloMo header

An implementation of video frame interpolation as described in the wildy popular research of Jing Et al. using an end-to-end convolutional neural network for generating spatially and temporally coherent video sequences

The main idea is to compute bi-directional optical flows from two frames using a U-Net architecture and combining them to approximate immediate flows

The data was trained on the adobe240fps dataset