Few-Shot Object Detection Based on the Transformer and High-Resolution Network

Now object detection based on deep learning tries different strategies. It uses fewer data training networks to achieve the effect of large dataset training. However, existing methods usually do not balance between network parameters and data. makes information provided by a small amount picture insufficient optimize model parameters, resulting in unsatisfactory results. To improve accuracy few shot detection, this paper proposes transformer high-resolution feature extraction (THR). High-resolution maintains resolution representation image. Channels spatial attention are used make focus features that more useful object. In addition, recently popular is fuse This compensates for previous failure making full use features. Experiments Pascal VOC MS-COCO datasets prove THR has achieved better results than mainstream detection.