Channel-Wise Pruning via Learn Gates&BN for Object Detection Tasks

Min Xiang Chen, Po Han Chen, Chia Chi Tsai

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Network pruning is an important research area aimed at addressing the high computational costs of deep neural networks. Previous studies [1] have indicated that it is not necessary to follow the conventional pruning process of training a large, redundant network, but rather, more diverse and higher-performing potential models can be directly pruned from randomly initialized weights. However, our experimental results on the MS COCO 2017 dataset [2] demonstrate that this approach is not applicable when compressing object detection models. We believe this outcome is associated with the complex network structures involved in object detection, making it relatively challenging to explore pruned architectures from random weights. To address this issue, we improved the existing Learn gates method and incorporated Batch normalization [3] to jointly learn channel importance. This enhances the learning capability of channel importance in a shorter time frame and facilitates the exploration of suitable pruned network structures within pre-trained weights. When applying our network pruning method to object detection models YOLOv3 [4] and YOLOv4 [5], our approach achieves higher accuracy with only a brief period of network structure learning.

Original languageEnglish
Title of host publicationGenetic and Evolutionary Computing - Proceedings of the Fifteenth International Conference on Genetic and Evolutionary Computing, 2023
EditorsJeng-Shyang Pan, Zhigeng Pan, Pei Hu, Jerry Chun-Wei Lin
PublisherSpringer Science and Business Media Deutschland GmbH
Pages238-249
Number of pages12
ISBN (Print)9789819994113
DOIs
Publication statusPublished - 2024
Event15th International Conference on Genetic and Evolutionary Computing, ICGEC 2023 - Kaohsiung, Taiwan
Duration: 2023 Oct 62023 Oct 8

Publication series

NameLecture Notes in Electrical Engineering
Volume1114 LNEE
ISSN (Print)1876-1100
ISSN (Electronic)1876-1119

Conference

Conference15th International Conference on Genetic and Evolutionary Computing, ICGEC 2023
Country/TerritoryTaiwan
CityKaohsiung
Period23-10-0623-10-08

All Science Journal Classification (ASJC) codes

  • Industrial and Manufacturing Engineering

Fingerprint

Dive into the research topics of 'Channel-Wise Pruning via Learn Gates&BN for Object Detection Tasks'. Together they form a unique fingerprint.

Cite this