MaGGIe Architecture: Efficient Mask-Guided Instance Matting

MaGGIe introduces an efficient framework using Cross-Attention, Self-Attention, and Sparse Convolutions for mask-guided instance matting, ensuring high accuracy and low latency.