The implementation of landmark-based, selective motion magnification can be found here. The sparse PCA-based decomposition and demo has been implemented by @cosmas-heiss and the RAFT training on faces for micro-expressions by @phflot.

Setup anaconda via conda env create -f environment.yml. To reproduce the demo recording, the file EP04_04f.avi from the CASME II microexpression dataset is required.

If you use this code for your work, please cite

P. Flotho, C. Heiss, G. Steidl, and D. J. Strauss, “Lagrangian Motion Magnification with Double Sparse Optical Flow Decomposition,” arXiv preprint arXiv:2204.07636 (2022).

BibTeX entry

@article{flotea2022c,
  doi = {10.48550/ARXIV.2204.07636},  
  author = {Flotho, P. and Heiss, C. and Steidl, G. and Strauss, D. J.},
  title = {Lagrangian Motion Magnification with Double Sparse Optical Flow Decomposition},
  journal = {arXiv},
  year = {2022},
  copyright = {Creative Commons Zero v1.0 Universal},
}