Temporally Coherent Video Harmonization Using Adversarial Networks

TL;DR

This paper introduces a novel adversarial network-based method for video harmonization that enhances realism and temporal consistency without requiring foreground masks, trained on a synthetic dataset that generalizes well to real-world videos.

Contribution

It proposes a new adversarial framework with a pixel-wise disharmony discriminator and a temporal loss for improved video harmonization, eliminating the need for input masks.

Findings

Achieves more realistic and temporally consistent video harmonization.

Successfully trained on a synthetic dataset that generalizes to real-world videos.

Outperforms previous methods in harmonization quality.

Abstract

Compositing is one of the most important editing operations for images and videos. The process of improving the realism of composite results is often called harmonization. Previous approaches for harmonization mainly focus on images. In this work, we take one step further to attack the problem of video harmonization. Specifically, we train a convolutional neural network in an adversarial way, exploiting a pixel-wise disharmony discriminator to achieve more realistic harmonized results and introducing a temporal loss to increase temporal consistency between consecutive harmonized frames. Thanks to the pixel-wise disharmony discriminator, we are also able to relieve the need of input foreground masks. Since existing video datasets which have ground-truth foreground masks and optical flows are not sufficiently large, we propose a simple yet efficient method to build up a synthetic dataset supporting supervised training of the proposed adversarial network. Experiments show that training on our synthetic dataset generalizes well to the real-world composite dataset. Also, our method successfully incorporates temporal consistency during training and achieves more harmonious results than previous methods.