search

HRNet / HRNet-MaskRCNN-Benchmark

Object detection with multi-level representations generated from deep high-resolution representation learning (HRNetV2h).
MIT License
138 stars 37 forks source link

higherhrnet？ #5

Closed henbucuoshanghai closed 4 years ago

henbucuoshanghai commented 4 years ago

higherhrnet？ the last stage featuremap aggregation???it means what??

aggregation it means add? or featuremap concate????

and tagmaps it means???? I dont konw tagmap

wondervictor commented 4 years ago

Hi, HigherHRNet is here Higher-HRNet-Human-Pose-Estimation

henbucuoshanghai commented 4 years ago

I have read the paper,but I dont understand sth.

henbucuoshanghai commented 4 years ago

the last stage featuremap aggregation???it means what??

aggregation it means add? or featuremap concate????

and tagmaps it means???? I dont konw tagmap

wondervictor commented 4 years ago

@henbucuoshanghai This repo is HRNet for Object Detection. As for feature aggregation in HRNetV2p, we use concatenation. What's tagmaps/tagmap?

henbucuoshanghai commented 4 years ago

Tagmaps are trained differently from heatmaps in High-erHRNet. We only predict tagmaps at the lowest resolution,instead of using all resolutions. This is because learningtagmaps requires global reasoning and it is more suitable topredict tagmaps in lower resolution. Empirically, we alsofind higher resolutions do not learn to predict tagmaps welland even do not converge. Thus, we follow [22] to train thetagmaps on feature maps at14resolution of input image

henbucuoshanghai commented 4 years ago

Tagmaps it means???????

wondervictor commented 4 years ago

@henbucuoshanghai Your problem has nothing to do with this repo. This repo is for Object Detection not Pose Estimation

henbucuoshanghai commented 4 years ago

ok i see,thanks