From 37a066327e41a160182fd250560f0290090575a3 Mon Sep 17 00:00:00 2001 From: daquexian Date: Sun, 5 May 2019 13:50:43 +0800 Subject: [PATCH] Update README for a more decent description of our speed --- README.md | 2 +- README_CN.md | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) diff --git a/README.md b/README.md index 2ec4842..ca1fd84 100644 --- a/README.md +++ b/README.md @@ -14,7 +14,7 @@ QQ Group (Chinese): 1021964010, answer: nndab Binary neural networks have great potential on edge devices since they replace float operations by efficient bit-wise operations. However, to leverage the efficiency of bit-wise operations, the reimplmentation of convolution layer and also other layers is needed. -To our best knowledge, dabnn is the first highly-optimized binary neural networks inference framework for mobile platform. We implemented binary convolutions with armv8 assembly. On Google Pixel 1, our dabnn is **700%~2300% faster** than [BMXNet](https://github.com/hpi-xnor/BMXNet) on a single binary convolution, and about **600% faster** than it on binarized ResNet-18. +To our best knowledge, dabnn is the first highly-optimized binary neural networks inference framework for mobile platform. We implemented binary convolutions with armv8 assembly. On Google Pixel 1, our dabnn is as **800%~2400% faster** as [BMXNet](https://github.com/hpi-xnor/BMXNet) on a single binary convolution, and as about **700% faster** as it on binarized ResNet-18. ## Benchmark and Comparison diff --git a/README_CN.md b/README_CN.md index de79b0b..8ee6c58 100644 --- a/README_CN.md +++ b/README_CN.md @@ -14,7 +14,7 @@ QQ 群：1021964010, 入群答案: nndab 然而，想发挥出位运算的高效率，就需要用位运算对卷积进行重新实现。一直以来，始终没有人针对二值网络推出一个高度优化的 inference 框架，这让二值网络令人遗憾的失去了可以部署在现有通用设备上这个巨大的优势。 -为了填补这个巨大的空白，我们推出了 dabnn，一个用 armv8 汇编重写了卷积，高度优化的二值网络 inference 框架。实验显示 dabnn 相对于现有的二值网络和浮点网络 inference 框架均能带来极大的加速。我们希望 dabnn 的开源可以成为二值网络部署在终端设备的关键一步，也可以有助于使这一领域的爱好者了解二值网络在真实设备上的表现。 +为了填补这个巨大的空白，我们推出了 dabnn，一个用 armv8 汇编重写了卷积，高度优化的二值网络 inference 框架。实验显示 dabnn 相对于现有的二值网络和浮点网络 inference 框架均能带来极大的加速。在运行单个二值卷积时，我们的速度是 BMXNet 的 **800%~2400%**，在运行二值化的 ResNet-18 时，我们的速度是 BMXNet 的约 **700%**。我们希望 dabnn 的开源可以成为二值网络部署在终端设备的关键一步，也可以有助于使这一领域的爱好者了解二值网络在真实设备上的表现。 ## 速度