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

about the infer fps and the network #9

Closed accountcwd closed 6 years ago

accountcwd commented 6 years ago

HI,i head that your implement run 23fps in GTX1080Ti .Did you use mobilenetV2 or some other lite architecture to achieve 23fps?Or you just use traditional convolution which means it can be faster using mobilenetV2?

mkocabas commented 6 years ago

We use a ResNet-50 to get that performance on GTX-1080ti. You can use mobilenet or variants to get higher FPS.

ZhongXiaoFang commented 6 years ago

Hi! thank you for you excellent job. When I trained on COCO dataset, the annotations are placed in /data/annotations/. but which path should the COCO images be placed ? @mkocabas