Paper1:《Instance-aware Semantic Segmentation via Multi-task Network Cascades》
本论文为CVPR16 arxiv:1512 Micorsoft Research
Author：Jifeng Dai Kaiming He Jian Sun
(1)提出Muti-task Network Cascades(多任务网络级联，MNC)，由differentiating instances、estimating masks和categorizing object组成(三者共享卷积)。
(2)single-step training、end-to-end training。在VOC 12 Instance segmentation task取得了new state-of-the-art performance.此外在object detection 任务上也超越了Faster RCNN
(3)提出了RoI Warping pooling
(4)MS COCO 15 segmentation第一名。
Paper2:《Fully Convolutional Instance-aware Semantic Segmentation》
本论文为CVPR17 arxiv:1611 Microsoft Research Asia
Author:Jifeng Dai等
(1)the first fully convolutional end-to-end solution for instance-aware semantic segmentation task.
(2)性能超越了MNC( COCO 15第一名)，以很大优势赢得了COCO 16 segmentation比赛第一名。
Paper3:《Mask R-CNN》
本论文为ICCV17 arxiv:1703 FAIR
Author: Kaiming He Georgia Gkioxari Piotr Dollar Ross Girshick
(1)extending Faster R-CNN by adding a branch for predicting an object mask in parallel with the existing branch for bounding box recognition.
(2)proposing a simple,quantization-free layer,called RoIAlign,that preserves exact spatial locations.
在Fast/Faster R-CNN中RoI pooling直接对子窗口采用max-pooling.we propose an RoIAlign layer that removes the harsh quantization of RoIPool, properly aligning the extracted features with the input. Our proposed change is simple: we avoid any quantization of the RoI boundaries or bins (i.e., we use x=16 instead of [x=16]). We use bilinear interpolation [22] to compute the exact values of the input features at four regularly sampled locations in each RoI bin, and aggregate the result (using max or average).
(3)Mask R-CNN outperforms all existing, single-model entries on every task, including the COCO 2016 challenge winners(FCIS).