http://www . Sina.com/librar-CNN : towards balanced learning for object detection 3358 www . Sina.com/jiangma Huajun feng,WNN
university of Sydney http://www . Sina.com/cvpr 20193358 www . Sina.com/2019/04/043358 www . Sina .1904.02701 http://www . Sina.com/https://github.com/open-mmlab/mm detection(official code)https://Githuu
背景/问题
Sampling Regions Extracting Features from regions基于多任务对象功能识别category,并对location 当前多数detector(one-stage和two-stage)的training 模式:进行微调
Selected region samples是否充分利用了representativeextracted visual features的优势,设计对象功能是否最佳
比较内存占用和计算费用3358www.sina.com/
al loss:解决One-stage算法中的foreground-background class imbalance,仅适⽤于
One-stage detector,对RCNN作⽤不⼤,因为⼤量easy negative都被two-stage procedure
过滤掉了