Efficient Architecture Search for Real-Time Instance Segmentation
Renqiu Xia, Dongyuan Zhang, Yixin Dong, Juanping Zhao, Wenlong Liao, Tao He, Junchi Yan
Abstract
Traditional CNN-based training for instance segmentation is time-consuming owing to large datasets and complex network modules, making direct searching of architecture challenging. In this paper, we introduce an efficient framework, named EASInst. It can discover practical backbone and encoder architectures for the improved sparse activation instance segmentation model. Specifically, we construct a supernet for both backbone and encoder modules of SparseInst based on a differentiable method. In addition, kernel sharing mask and channel pruning technology are employed. Moreover, Taylor-Loss and a novel DY-Loss are devised for instance segmentation to improve the accuracy. Experiments show that the searched architectures outperform the existing Resnet-based real-time instance segmentation methods, which achieve 38.5 mAP with 39.5 FPS on COCO test-dev set.
BibTeX
@inproceedings{icassp2024_efficientarchite,
title = {Efficient Architecture Search for Real-Time Instance Segmentation},
author = {Renqiu Xia and Dongyuan Zhang and Yixin Dong and Juanping Zhao and Wenlong Liao and Tao He and Junchi Yan},
booktitle = {ICASSP 2024},
year = {2024}
}