search

lawy623 / LSTM_Pose_Machines

Code repo for "LSTM Pose Machines" (CVPR'18)
276 stars 43 forks source link

Some questions about paper details #3

Open FightForCS opened 6 years ago

FightForCS commented 6 years ago

Hi, thanks for your great paper and code. I have some questions:

  1. How long does it take to pretrain on LEEDS Sport Dataset and MPII Dataset ? And How long does it take to train on PENN dataset after pretraining?

  2. I notice the network does not use feature extractor which is pretrained on ImageNet. Why not? Since it might speed up training, as CPM author stated in https://github.com/shihenw/convolutional-pose-machines-release/issues/38

  3. How about the performance if we only train on PENN dataset, without pretraining on LEEDS Sport Dataset and MPII Dataset? Will the performance drop a lot ?

Thanks.

pasawaya commented 6 years ago

Not one of the authors but I wrote a Pytorch implementation and got up to 86% PCK@0.2 accuracy in 24 hours when pre-training on MPII.