What is the goal for aligning the keypoints?

liruilong940607 / Pose2Seg

Code for the paper "Pose2Seg: Detection Free Human Instance Segmentation" @ CVPR2019.

http://www.liruilong.cn/projects/pose2seg/index.html

MIT License

532 stars 136 forks source link

What is the goal for aligning the keypoints? #4

Closed zzw1123 closed 5 years ago

zzw1123 commented 5 years ago

Hi, thanks for your contribution! I have a question that why should we align all the keypoints? For ROI align, the features must be input to fully connected layer so the size of them should be the same. But in the segmentation task, there is no fc layer and I cannot figure out the function of affine-align operation.

liruilong940607 commented 5 years ago

Thanks for your attention!

An alignment brings several benefits: (1) Human in the image have wide range of size. An aligned size and centralisation makes segmentation network easy to learn. (2) Alignment reduce the calculation. Otherwise, to do instance segmentation, we will need to process whole image for each instance. It will cost a lot more gpu memory as well as calculation.