Table 2 Comparisons of different methods on the proposed dataset
From: Classroom student posture recognition based on an improved high-resolution network
Method | Reading (%) | Looking (%) | Hunching (%) | Accuracy (%) |
|---|---|---|---|---|
OpenPose + SVM | 67.3 | 61.4 | – | 64.2 |
Faster RCNN + HRNet-W32 + SVM | 83.4 | 84.5 | 92.4 | 81.8 |
YOLOV3 + SE-HRNet-W32 + SVM (our) | 88.6 | 89.2 | 91.6 | 90.1 |