search

houqb / SPNet

Code for our CVPR2020 paper "Strip Pooling: Rethinking Spatial Pooling for Scene Parsing"
MIT License
387 stars 57 forks source link

About the SPM #7

Closed yexiguafuqihao closed 4 years ago

yexiguafuqihao commented 4 years ago

It is a very nice job. However, I can't find the schematic illustration strip pooling module implementation in this codebase, which is depicted in Figure.2 in the paper. Would you please provide me an explanation ?

houqb commented 4 years ago

Check it out in the models/resnet.py

Kind regards, Qibin (Andrew)

On Thu, May 7, 2020 at 1:33 PM yexiguafuqihao notifications@github.com wrote:

It is a very nice job. However, I can't find the schematic illustration strip pooling module implementation in this codebase, which is depicted in Figure.2 in the paper. Would you please provide me an explaination?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/Andrew-Qibin/SPNet/issues/7, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADNCN44NCU35TLP5K726V4TRQJB3BANCNFSM4M3AQPKQ .

swjtulinxi commented 4 years ago

actually，The author's model accuracy improvement mainly comes from SPM, not MPM. If we can design complex SPM combination in MPM stage, we should also be able to improve the accuracy

swjtulinxi commented 4 years ago

事实上，大佬，你给我们提供的一个新思路就是，对于语义分割来说，目前的后端模块想ppm，aspp，和注意力机制等，其实性能已经足够好了，其实最主要的是，我们后端需要接收来自前端网络的信息，其实前端网络提取的信息往往比后端更重要，新思路就是目前我们可以在前端做工作，因为对后端的改进似乎已经很难了，而且效果也不是很强了