Jan, 2024
CoordGate: 在卷积神经网络中高效地计算空间变化卷积
CoordGate: Efficiently Computing Spatially-Varying Convolutions in Convolutional Neural Networks
Sunny Howard, Peter Norreys, Andreas Döpp
TL;DROptical imaging systems often face limitations in resolution due to the point spread function, but CoordGate, a lightweight module using a multiplicative gate and coordinate encoding network, offers an efficient solution for spatially-varying convolutions in CNNs, outperforming conventional approaches in image deblurring.