BoundaryNet

BoundaryNet is an innovative resizing-free approach used to annotate layouts for images. This approach utilizes a variable-sized region of interest, which is first entered into an attention-guided skip network. This network is then optimized via Fast Marching distance maps to provide an initial estimate of the boundary and an associated feature representation. Finally, these outputs are processed through a Residual Graph Convolution Network, which is optimized using Hausdorff loss, to produce th