Why restore resolution through deconv instead of HRNet?

HRNet / HigherHRNet-Human-Pose-Estimation

This is an official implementation of our CVPR 2020 paper "HigherHRNet: Scale-Aware Representation Learning for Bottom-Up Human Pose Estimation" (https://arxiv.org/abs/1908.10357)

MIT License

1.35k stars 272 forks source link

Why restore resolution through deconv instead of HRNet? #41

Closed jialeli1 closed 4 years ago

jialeli1 commented 4 years ago

Compared to HRNet, the newly added deconv operation can restore the resolution again. But why not directly use 1/2 resolution in the first stage of HRNet, so that the final 1/2 resolution can be directly restored in the style of HRNet. Is there any limitation or deficiency? This really confuses me, thank you for your answer.

dangnh0611 commented 1 year ago

@jialeli1 hi, could you share your thoughts about this ? refer: https://github.com/HRNet/HigherHRNet-Human-Pose-Estimation/issues/19