Transformers can adaptivelyaggregate similar features from a global view using self-attention mechanism . The CA-FPN significantly outperforms competitive baselines without bells and whistles . Code will be made publicly available and code will be publicly available . For object detection, Feature Pyramid Network (FPN) proposes featureinteraction across layers and proves its extremely importance. However, itsinteraction is still in a local manner, which leaves a lot of room for improvement. What’s more, light transformers can further make theapplication of multi-head attention mechanism easier. Most importantly, ourCA-FPNs can be readily plugged into existing FPN-based models. Extensiveexperiments on the challenging COCO object detection dataset demonstrated thatour CA- FPN significantly outranks competitive basinals without bells &whistles. Without bells and whistle

Author(s) : Yongxiang Gu, Xiaolin Qin, Yuncong Peng, Lu Li

Links : PDF - Abstract

Code :

https://github.com/oktantod/RoboND-DeepLearning-Project


Coursera

Keywords : fpn - transformers - bells - object - network -

Leave a Reply

Your email address will not be published. Required fields are marked *