Pose-Guided High-Resolution Appearance Transfer via Progressive Training

(paper ID 968)

Supplementary Material

1024 x 1024 Test Results

Given a reference image (leftmost column) and target poses (top row), APT transfers the target pose seamlessly to the reference person in high resolution. Click on each sub-image to view full-sized image.

YouTube dataset


Human3.6M dataset

Pose-Guided Human Video Generation

In this section we present an application of APT on Pose-Guided Human Video Generation

1. We compare our generation results with Cai et al (Deep video generation, prediction and completion of human action sequences, ECCV 2018)

2. We present additional results on Pose-Guided Human Video Generation

Disclaimer: Our videos are generated frame-by-frame, unlike Cai et al which consider temporal coherence. Yet, our 1024 x 1024 video demonstrates significantly higher quality.

Comparison with Cai et al

Target Pose

Cai et al

Ours


Additional results

Target Pose

Ground Truth

Our result