search

xingyizhou / pytorch-pose-hg-3d

PyTorch implementation for 3D human pose estimation
GNU General Public License v3.0
615 stars 141 forks source link

What's the meaning of 'scale' in mpii dataset? #2

Closed EthanGuan closed 6 years ago

EthanGuan commented 6 years ago

if I use my own dataset, every image is 640 * 480, what's the value of scale should I set it to?

xingyizhou commented 6 years ago

Hi EthanGuan, Thanks for your interest in this project. The 'scale' in the mpii dataset is provided as an annotation, it is the square-shape human bounding size. It is dependent on the human bounding box size, not just the input image size. But it is safe to just set the scale to 640 and center at (320, 240).

bwuzhang commented 6 years ago

Would you please elaborate on this? How do I get bounding box given center and scale?

xingyizhou commented 6 years ago

The upper-left and bottom-right coordinates of bounding box are (center[0] - scale / 2, center[1] - scale / 2), (center[0] + scale / 2, center[1] + scale / 2).

bwuzhang commented 6 years ago

I guess by saying 'scale' you mean '200 * scale'. I did try this but I got some absurd annotation...

On Feb 28, 2018 9:14 PM, "Xingyi Zhou" notifications@github.com wrote:

The upper-left and bottom-right coordinates of bounding box are (center[0]

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/xingyizhou/pytorch-pose-hg-3d/issues/2#issuecomment-369451278, or mute the thread https://github.com/notifications/unsubscribe-auth/AIhx7ze-JgPb9UpXBgs3v6-vAImboFYkks5tZ1lzgaJpZM4RMouo .

xingyizhou commented 6 years ago

Yes, it is '200 x scale' based on Newell's hourglass implementation. What's the following question?