期刊
COMPUTER VISION, ECCV 2022, PT XII
卷 13672, 期 -, 页码 17-33出版社
SPRINGER INTERNATIONAL PUBLISHING AG
DOI: 10.1007/978-3-031-19775-8_2
关键词
Mobile network; Quantization; Neural architecture search
资金
- NRF - Korea government (MSIT) [2021-0-00105, NRF-2021M3F3A2A02037893]
- Samsung Electronics (Memory Division, SAIT) [SRFC-TC1603-04]
- IITP
In this paper, a new method for low-bit activation quantization called BASQ is proposed. A novel block structure suitable for both MobileNet and ResNet structures is also introduced. The proposed method offers competitiveness across various low precisions, outperforming existing methods in terms of accuracy on ImageNet.
In this paper, we propose Branch-wise Activation-clipping Search Quantization (BASQ), which is a novel quantization method for low-bit activation. BASQ optimizes clip value in continuous search space while simultaneously searching L2 decay weight factor for updating clip value in discrete search space. We also propose a novel block structure for low precision that works properly on both MobileNet and ResNet structures with branch-wise searching. We evaluate the proposed methods by quantizing both weights and activations to 4-bit or lower. Contrary to the existing methods which are effective only for redundant networks, e.g., ResNet-18, or highly optimized networks, e.g., MobileNet-v2, our proposed method offers constant competitiveness on both types of networks across low precisions from 2 to 4-bits. Specifically, our 2-bit MobileNet-v2 offers top-1 accuracy of 64.71% on ImageNet, outperforming the existing method by a large margin (2.8%), and our 4-bit MobileNet-v2 gives 71.98% which is comparable to the full-precision accuracy 71.88% while our uniform quantization method offers comparable accuracy of 2-bit ResNet-18 to the state-of-the-art non-uniform quantization method. Source code is on https://github.com/HanByulKim/BASQ.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据