Some questions about the different configs in the "specify_parameter" function in the cpm_vgg16.py

facebookresearch / supervision-by-registration

Supervision-by-Registration: An Unsupervised Approach to Improve the Precision of Facial Landmark Detectors

Other

761 stars 165 forks source link

Some questions about the different configs in the "specify_parameter" function in the cpm_vgg16.py #35

Closed SunnyDp closed 5 years ago

SunnyDp commented 5 years ago

Thanks to your great job! I have some problems about the the “param_dict” variable in the "specify_parameter" function. The function aims to get the parameters from the model, right? But it has some different parameter settings, such as different lr and bias=false/true...Why one model has many different param settings?

D-X-Y commented 5 years ago

specify_parameter aims to make different layers to have different learning rates. Since our base detector is convolutional pose machine (CPM), we follow their training setup (https://github.com/shihenw/convolutional-pose-machines-release/blob/master/training/genProto.py#L62), which uses different LR for different layers.