TensorFlow 加载模型前n-1层

最新推荐文章于 2022-03-11 15:41:55 发布

yan__csdn

最新推荐文章于 2022-03-11 15:41:55 发布

阅读量1.5k

点赞数

分类专栏：代码

本文链接：https://blog.csdn.net/qq_40614981/article/details/83017105

版权

      path_checkpointa=r'E:\anacondaCode\ResNet_cifar-master181010(2000) - 副本\data\cifar10_tf\checkpoint_'
      ckpt = tf.train.get_checkpoint_state(path_checkpointa)#模型A的checkpoint
      if ckpt and ckpt.model_checkpoint_path:
          print('restore success ')
          
          saver.restore(sess,ckpt.model_checkpoint_path)
          
          var_list = [['logit/DW'] ,['logit/DW/Momentum '],['logit/biases'],['logit/biases/Momentum '],modelB.global_step]
          initfc = tf.variables_initializer(var_list, name='init')        
          sess.run(initfc)
#加载整个模型 然后初始化部分层的参数

#      #加载模型 除Logit之外
      path_checkpointa=r'E:\anacondaCode\ResNet_cifar-master181010(2000) - 副本\data\cifar10_tf\checkpoint_'
      ckpt = tf.train.get_checkpoint_state(path_checkpointa)
      if ckpt and ckpt.model_checkpoint_path:


          variables &