Efficient Accelerator for Dilated and Transposed Convolution with Decomposition
arXiv:2205.02103 · doi:10.1109/ISCAS45731.2020.9180402
Abstract
Hardware acceleration for dilated and transposed convolution enables real time execution of related tasks like segmentation, but current designs are specific for these convolutional types or suffer from complex control for reconfigurable designs. This paper presents a design that decomposes input or weight for dilated and transposed convolutions respectively to skip redundant computations and thus executes efficiently on existing dense CNN hardware as well. The proposed architecture can cut down 87.8\% of the cycle counts to achieve 8.2X speedup over a naive execution for the ENet case.
arXiv comment10 pages, 12 figures, published in IEEE ISCAS 2020
References in corpus (5)
- Very Deep Convolutional Networks for Large-Scale Image Recognition
- MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications
- ENet: A Deep Neural Network Architecture for Real-Time Semantic Segmentation
- DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs
- VWA: Hardware Efficient Vectorwise Accelerator for Convolutional Neural Network