Latent Image Animator — Reimplementation

Reproducibility work on a self-supervised motion-transfer method: animating a still image by applying latent-space linear transformations learned from a driving video, with no paired supervision.

The project reimplemented the method from the paper, reproduced the full training pipeline, and evaluated transfer fidelity and temporal coherence on standard video benchmarks — a useful exercise in how much of a published result survives independent reimplementation.

Stack: PyTorch, OpenCV, NumPy, self-supervised learning

Code: github.com/keniel123/latent-image-animator