训练FCN-VGG16导入之前训练好的npy文件，出现No gradients for any variables

出现如下错误：

Traceback (most recent call last):
  File "/data2/test/cxj/fcn_vgg16/test_fcn16_vgg.py", line 134, in <module>
    loss, optimizer, fcn_prob,fcn_pred,fcn_pred_up,lr = train_net(vgg_fcn=fcn, input_tensor=images, out_tensor=true_out)
  File "/data2/test/cxj/fcn_vgg16/test_fcn16_vgg.py", line 72, in train_net
    apply_gradient_op = optimizer.apply_gradients(grads)
  File "/home/test/anaconda3/envs/mask_rcnn/lib/python3.4/site-packages/tensorflow/python/training/optimizer.py", line 591, in apply_gradients
    ([str(v) for _, v, _ in converted_grads_and_vars],))
ValueError: No gradients provided for any variable: ["<tf.Variable 'upscore2/up_filter:0' shape=(4, 4, 2, 2) dtype=float32_ref>", "<tf.Variable 'upscore32/up_filter:0' shape=(32, 32, 2, 2) dtype=float32_ref>"].

 

后面经过验证，是因为将原始的网络定义模型中的trainable这个关键字删除了，导致原来的npy文件无法正确导入进去。