Home

Awesome

MoTrans: Customized Motion Transfer with Text-driven Video

[Paper]

News

<img src="imgs/pipeline.png" width="100%"> <img src="imgs/qualitative.png" width="100%">

Dataset

We collect a dataset that includes 12 distinct motion patterns, sourced from the Internet, the UCF101 dataset, the UCF Sports Action dataset, and NTU RGB+D 120. Each motion pattern is represented by approximately 4-10 training videos. The dataset consists of various sports motions, such as weightlifting and golf swing, alongside large-scale limb movements like waving hands and drinking water.

TODOs