Created by: busyboxs

机器环境：ubuntu16.04 + CUDA8 + cudnn5.1 + paddlepaddle1.1（post85）+ anaconda(py2.7)

如果不改类别数目，导入预训练模型可以正常运行；

如果修改类别数目，不导入预训练模型可以正常运行；

如果修改类别数目，导入预训练模型则会报错；

错误为

Traceback (most recent call last):
  File "./train.py", line 169, in <module>
    fetch_list=[pred, loss_mean])
  File "/home/yangshun/anaconda2/envs/paddle1.2/lib/python2.7/site-packages/paddle/fluid/executor.py", line 470, in run
    self.executor.run(program.desc, scope, 0, True, True)
paddle.fluid.core.EnforceNotMet: Enforce failed. Expected in_dims.size() == filter_dims.size(), but received in_dims.size():4 != filter_dims.size():1.
Conv input dimension and filter dimension should be the same. at [/paddle/paddle/fluid/operators/conv_op.cc:50]

很奇怪，调试半天也不知道为什么，