arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2205.02103cs.ARcs.LG
收藏

一种有效的带分解的扩张和转置卷积加速器

Efficient Accelerator for Dilated and Transposed Convolution with Decomposition

National Chiao Tung University(国立交通大学)

Kuo-Wei Chang, Tian-Sheuan Chang

摘要:

用于扩张和转置卷积的硬件加速使得能够实时的执行相关任务,如分割,但是当前的设计是特定于这些卷积类型的,或者遭受用于可重构设计的复杂控制。本文提出了一种设计,分别为扩张卷积和转置卷积分解输入或权重,以跳过冗余计算,从而在现有的密集CNN硬件上有效地执行。该架构可以减少87.8%的周期数,实现8.2倍的加速比在一个天真的执行ENet的情况下。

英文摘要:

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.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑