search

Daniil-Osokin / gccpm-look-into-person-cvpr19.pytorch

Fast and accurate single-person pose estimation, ranked 10th at CVPR'19 LIP challenge. Contains implementation of "Global Context for Convolutional Pose Machines" paper.
Apache License 2.0
168 stars 28 forks source link

Change backbone network,Will it well？ #13

Closed Willert98 closed 4 years ago

Willert98 commented 4 years ago

Thank for your share ,I noticed that the backbone is mobilenet v1 , Have you try change backbone network,Will it well?

Willert98 commented 4 years ago

emmm,And can you share your work Validation....information ,such as Head 80 .09 Sho 70.0...

Daniil-Osokin commented 4 years ago

Hi, we did comparison with MobileNet V2 for the similar project. For some reasons it did not perform well, so we use MobileNet V1 as feature extractor. Mind running validation to check the numbers or you may look into the paper.

Daniil-Osokin commented 4 years ago

Hope, it helped.