MaGGIe improves accuracy and efficiency in instance matting via transformer attention and sparse convolution, with future goals in weakly-supervised learning. - https://t.co/f0qGSCZjZU #deeplearning #maggiediscussion