Year: 2020
Type: conference-paper
Abstract: With the rapid development of deep learning image recognition technology, people have higher and higher requirements on the accuracy and speed of segmentation recognition. So instance segmentation becomes more and more important. The accuracy and speed of segmentation recognition are increasingly demanded. Recently, Shen Chunhua et al. proposed a very quantized instance segmentation model CondInst. In the paper, which used CSPResNet instead of ResNet, and added a Prediction module method to the head to improve model accuracy. Finally, on the MS COCO2017 data set, the simulation results show that CondInst’s AP, AP50, AP75, APS, APM and APL were increased by 0.2%, 0.4%, 0.3%, 0.1% and 0.4% respectively in the 50 layer residual model of our model, compared with CondInst’s AP, AP50, AP75, APS, APM and APL.The experimental results show that the method has certain practical value.
Authors: Jianye Yuan, Xinwang ding, Lele Sun
Institutions: Xinjiang University, Shandong Huayu University of Technology
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W3108970668