search

mkocabas / pose-residual-network

Code for the Pose Residual Network introduced in 'MultiPoseNet: Fast Multi-Person Pose Estimation using Pose Residual Network (ECCV 2018)' paper
https://arxiv.org/abs/1807.04067
346 stars 92 forks source link

A question abount performance on embedded systems #6

Closed IgorMunizS closed 6 years ago

IgorMunizS commented 6 years ago

Hi, great work! I am researching the application of person detection and pose estimation in robots and I'm wondering if you tested the performance of the entire model on an embedded device like Jetson TX2. Do you think it's possible to replace RetinaNet with squeezeDet(https://arxiv.org/pdf/1612.01051.pdf) for detection?

Thanks!

mkocabas commented 6 years ago

I'm wondering if you tested the performance of the entire model on an embedded device like Jetson TX2.

No, we didn't run the model on an embedded device. We don't have any plan to do so in short tem.

Do you think it's possible to replace RetinaNet with squeezeDet(https://arxiv.org/pdf/1612.01051.pdf) for detection?

I don't know about squeezeDet model, however if it has a feature extractor part like in RetinaNet or Faster-RCNN, you can use it for sure.