The architercure of PyTorch implementation

xingyizhou / pytorch-pose-hg-3d

PyTorch implementation for 3D human pose estimation

GNU General Public License v3.0

614 stars 143 forks source link

The architercure of PyTorch implementation #67

Closed zhangyahu1 closed 5 years ago

zhangyahu1 commented 5 years ago

Hi Xingyi, I found the model architerture is different form your paper as you mentioned. Is it just one stack and one residual module followed by deconvoutional layers? and output 646416(joint heatmaps) and 646416(depth map)? Thanks!

xingyizhou commented 5 years ago

Hi, No, it is not the hourglass architecture anymore, but just resnet50 at conv5 (32x downsample) with 3x deconv upsample (back to 4x downsample). You can refer to the Simple baseline paper for details.

zhangyahu1 commented 5 years ago

Got it. Thanks for your answer!