In this short paper, we propose a baseline (off-the-shelf) for Continual
Learning of computer vision problems, by leveraging the power of pretrained
models. By doing so, we devise a simple approach achieving strong performance
for most of the common benchmarks. Our approach is fast sin