Details
Presenter(s)
Display Name
Peixiang Yang
- Affiliation
-
AffiliationNanjing University
- Country
Abstract
In this paper, we propose a fast reconfigurable scheme to accelerate different DeConvs. Then, based on the proposed scheme, we design a reconfigurable hardware architecture and an adaptive dataflow to handle several deconvolutional layers and convolutional layers flexibly. Finally, some deconvolutional models are chosen to evaluate our design. The experimental results show that our design can support more types of operations compared with previous works, and achieve promising computational efficiency.