Comments (12)
谢谢回复,
那么整个系统流程,按照我的理解,是
1)由某种方法(预测或者G.T)得到图像上multi-person的poses P
2)image过backbone network,得到feature map F
3)根据pose template和person pose,求H,将H应用到F上
4)根据P画出PAF和joint heatmaps
5)feature map + PAF + joint heatmap 过SegModule,得到图像上若干个人的Mask
6)Mask做AlignReverse得到最终结果
这样的理解对吗?
PS:Figure4最左的图(再加上image+pose as input这句话),对我而言有些许歧义,我原以为是:由joint生成heatmap,再和图像concat,过backbone
另外还有2个问题:
a)Figure4有两个 “+ Concat”,这两个有什么区别?是指不同的人吗?
b)我其实对segment工作并不是很了解,所以想请教一下。本文是做多人segment,假设图中有N个人,那么SegModule的输出直接是N个人的Mask吗?还是逐个输出每个人的Mask?
from pose2seg.
另外,http://www.liruilong.cn/Pose2Seg/index.html 这个网站404……
from pose2seg.
1.1)每个人pose的表示都是一个vector,见论文4.2.1。
1.2)论文Figure 4. 有写是concat。是按通道方向拼接。
2)Simple Baselines@MSRA是top-down的方法。top-down的方法由于NMS的存在对于遮挡问题从原理上无法处理。这部分的讨论和介绍见论文Introduction。
3)并没有预测两次。skeleton feature的生成不需要预测,就是直接用affine-align将pose vector坐标对齐后画出来的。
4)Figure4 是e2e的(input是image + pose)。
另,感谢提醒,网站连接已修复。
from pose2seg.
Right! You got the basic idea!
a) 没有区别,每个人都要concat。
b) 直接输出N个人的。输入是(N, C1, H, W)的feature,输出是(N, C2, H, W)的score map,对应于N个人。
from pose2seg.
你好,最后还有一个问题,
4.2.2提到要用BBOX去crop得到ROI,那么BBOX从哪里来?除了实验中提到可以用G.T,或者从GT KPT expand出来。
像是Figure7(a)的那些例子,bbox从哪里来的呢?
from pose2seg.
Figure 7(a) 中的不是bounding box,是affine-align对齐操作对应在原图上的region。
from pose2seg.
好的,谢谢,我基本明白了pipeline。
另外,Figure4的Affine-Align Operation给的feature map图,看上去是很像人体的热力图,换言之,它已经带有一定语义了。我想问的是,backbone出来的feature真的是长这样的吗?
from pose2seg.
feature map是256-channel的。Fig. 4 中我们为了可视化对256-channel做了个融合处理变成3-channel的heatmap。实际上,因为e2e地训练,backbone一定是能学到一些语义的。
from pose2seg.
明白,谢谢你的回复,祝你CVPR顺利!
from pose2seg.
你好!我也有几个问题:如果没有ground truthd的pose的话,比如表2里面的0.222这个,你的pose是从哪里来的?还有test上没有ground truth的pose,你怎么算模板呢?
from pose2seg.
from pose2seg.
还有一个问题,既然对于一张图都是全图预测多人,训练的时候人数不一致怎么办呢,concate的特征通道不一致怎么办?
from pose2seg.
Related Issues (20)
- The instance segmentation performance rely heavily on keypoints performance HOT 2
- coco 关键点修改 HOT 3
- Got wrong test results about AP(area=medium)
- Request the dataset address
- about the train error
- Can we give the network only the RGB image? If we don't have a pose model? HOT 1
- Testing with Openpose Keypoints HOT 2
- Combination with Human pose detection HOT 1
- Data Augmentation
- 测试的模型效果很好,如果想提升实例分割的速度,可以给一些建议吗?
- Some problems after using alphapose and then use pose2seg?
- What's the environment setup and requirement? HOT 1
- RuntimeError: CUDA out of memory in training HOT 1
- Training on smaller dataset HOT 1
- This repo need to have "inference.py/ipynb"
- Posture alignment
- pose point
- What is in this work's COCO dataset
- Reproducing results without GT Keypoints HOT 10
- Ski2DPose dataset
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from pose2seg.