Object Detection Method Based On Improved One-Stage Detector
Wei Hongtao, Xin Zhou Yang · 2020 5th International Conference on Smart Grid and Electrical Automation (ICSGEA) · 2020
We use an efficient way to simulate the two-stage detector, and optimize the algorithm to improve efficiency of object detection as the accuracy of the one-stage detector often lags behind that of the two-stage detector. We have improved the one-stage detector to get higher accuracy on SSD. We mainly adjust the default boxes so that it can well capture the range of the target object in the image, and then the module given the adjusted boxes responds to the alignment of the receptive field of the convolution filter. Test results on the PASCAL VOC dataset show that the final model accuracy can reach that of the advanced detector.