ECCV'18

Learning to forecast and refine residual motion for image-to-video generation

Long Zhao, Xi Peng, Mubbasir Kapadia, Dimitris Metaxas

European Conference on Computer Vision (ECCV), Munich, Germany, 2018

[Paper] [Code]

Green: target motion. Red: generate target motion given only the 1st frame of each clip.

1st row: Smile

2nd row: Disgust

3rd row: Surprise

Each clip (~30 frames) is generated given only the 1st frame.

1st row: Ground Truth

2nd row: Ours

3rd row: ICML'17

Quantized Densely Connected U-Nets for Efficient Landmark Localization

Zhiqiang Tang, Xi Peng, Shijie Geng, Lingfei Wu, Shaoting Zhang, Dimitris Metaxas

European Conference on Computer Vision (ECCV), Munich, Germany, 2018

[Paper] [Code]