pytorch冻结网络参数及网络层学习率设置

1、冻结网络层参数
参考：注意是requires_grad
https://zhuanlan.zhihu.com/p/357349586
https://zhuanlan.zhihu.com/p/65105409

 for name, param in model.named_parameters():
        if  ('discriminator_G' not in name) and ('discriminator_L' not in name):
            param.requires_grad = False

2.分层处理学习率
建立优化器的时候：

 	discriminator_L_param=list(map(id, model.backbone_2d.discriminator_L.parameters()))
    discriminator_G_param=list(map(id, model.backbone_2d.discriminator_G.parameters()))
    FL_param = list(map(id, model.backbone_2d.FL.parameters()))
    base_params = filter(lambda p: id(p) not in discriminator_L_param + discriminator_G_param+FL_param,model.parameters())

    if optim_cfg.OPTIMIZER == 'adam':
        optimizer = optim.Adam([
            {'params': base_params},
            {'params':model.backbone_2d.discriminator_G.parameters(),'lr':optim_cfg.LR*3},
            {'params':model.backbone_2d.discriminator_L.parameters(),'lr':optim_cfg.LR*3},
            {'params':model.backbone_2d.FL.parameters(),'lr':optim_cfg.LR*3}], 
            lr=optim_cfg.LR, weight_decay=optim_cfg.WEIGHT_DECAY)