四川通江县住房和建设局网站/江苏营销型网站建设
论文:https://arxiv.org/pdf/1803.01534.pdf
香港中文大学 2018 作品。PANet在COCO2017的实例分割上获得第一,在目标检测任务上获得第二。
作者通过研究Mask R-CNN发现:
(1)底层特征难以传达到高层次:features in low levels are helpful for large instance identification;But there is a long path from low-level structure to topmost features, increasing difficulty to access accurate localization information;作者设计了Bottom-up path augmentation。
(2)框是启发式生成,对最终预测有帮助的信息被丢弃:Further, each proposal is predicted based on feature grids pooled from one feature level, which is assigned heuristically.This process can be updated since information discarded in other levels may be helpful for final prediction。作者设计了Adaptive feature pooling。
(3)mask没有聚集多样化的信息去生成:Finally, mask prediction is made on a single view, losing the chance to gather more diverse information.作者设计了Fully-connected Fusion。
论文第一张图很重要,下图里的(b)是自下而上的路径增加 Bottom-up path augmentation,(c)是Adaptive feature pooling。红色线表达了图像底层特征在FPN中的传递路径,要经过100多层layers;绿色线表达了图像底层特征在PANnet 中的传递路径,只需要经过小于10层layers。
注意 Bottom-up path augmentation 里是add操作:
mask分支的细节(作者的这个叫Fully-connected Fusion,fc是位置敏感的 , 有利于区分不同实例,mask得更准):
box 分支的细节:
拉满的效果: