High Perspective Faster R-CNN
Download as PDF
DOI: 10.23977/CNCI2020022
Author(s)
Quankai Liu, Kefeng Li and Li Liu
Corresponding Author
Li Liu
ABSTRACT
In this paper the original Faster R-CNN model structure was improved to solve the problem of limited receptive fields. The RPN sub-module was combined with an attention layer to get a higher perspective, so that the new model can better analyze the whole picture information and highlight meaningful information. Experiments validate that our method achieves a new improvement for the object detection on PASCAL VOC2007 dataset. And when the change ratio of the middle multi-layer perceptron in the channel attention was set to 1, result reached the highest correct rate of 71%, more than the original model.
KEYWORDS
Faster R-CNN; attention; RPN; respective field