Caffe：手术后改变数据层

Question

我训练了一个带有HDF5数据层的FC网络，然后用手术移植到卷积网络，然后将数据层改为适合探测的数据层，即：

从：

layer {
    name: "layer_data_left"
    type: "HDF5Data"
    top: "data_left"
    top: "labels_left"
    include {
        phase: TRAIN
    }
    hdf5_data_param {
        source: "/home/me/Desktop/trainLeftPatches.txt"
        batch_size: 128
    }
}

到

layer {
  name: "data_left"
  type: "Input"
  top: "data_right"
  input_param { shape: { dim: 1 dim: 1 dim: 1241 dim: 367 } }
}

是否有任何原因导致内存不足？：

>>> fc_net.forward()
F0729 20:02:02.205382  6821 syncedmem.cpp:56] Check failed: error == cudaSuccess (2 vs. 0)  out of memory
*** Check failure stack trace: ***
Aborted (core dumped)

或者，我是否更有可能在外科手术中出错？交换数据层？

谢谢。