Fig. 1
From: Sparse self-attention aggregation networks for neural sequence slice interpolation
Overview of the SSAN algorithm, which includes the siamese residual dense network, attention-aware layers, and hybrid network. Given two input EM images, we first use the RDN module to calculate the forward and reverse features and then use the proposed attention-aware layer to generate warped intermediate frames. We then use a hybrid network adaptively fusing the warped intermediate frames to generate the final intermediate frame