una-dinosauria / 3d-pose-baseline

A simple baseline for 3d human pose estimation in tensorflow. Presented at ICCV 17.
MIT License
1.41k stars 355 forks source link

Inference time on Android? #189

Open serviceberry3 opened 3 years ago

serviceberry3 commented 3 years ago

Hi,

You say a feed-forward inference on this model is very fast--2ms on a Titan Xp. Do you think the speed would hold up on a recent Android device (like a Google Pixel 4)? I'm looking to do real-time inference using Tensorflow Posenet and this, and I'd like a speed of 20-30 FPS.

Thanks a lot! Great work.

serviceberry3 commented 3 years ago

Or do you know offhand whether Facebook's VideoPose3D would be slower or faster than your model? I tried running VideoPose3D on Android, but it's really slow; I'm wondering whether I should try your model as well...

una-dinosauria commented 3 years ago

Hi @serviceberry3 ,

Sorry, I don't know anything about inference on mobile devices. I'll leave the issue open in case someone else knows the answer.

Cheers,

zuozhen commented 2 years ago

Or do you know offhand whether Facebook's VideoPose3D would be slower or faster than your model? I tried running VideoPose3D on Android, but it's really slow; I'm wondering whether I should try your model as well...

Hi @serviceberry3 , Have you tried this on Android, please? Thanks.