3D Separable Winograd Convolutions with an Efficient Pipeline
Shen-Chin Chang, You-Sheng Xiao, Yu‐Cheng Fan · 2023
As time goes on, a neural network architecture becomes more and more complicated. Among of layers of the neural networks, the convolution layer accounts for a considerable proportion of the overall neural network operation, demonstrated by many pieces of literature. Therefore, this paper will focus on computing efficiency to improve the convolution layer, further optimizing the neural network.