How to store volumetric patch into HDF5?

问题

I have a volumetric data with size 256x128x256. Due to limit memory, I cannot use the whole data to directly feed to CAFFE. Hence, I will randomly choose n_sample patches 50x50x50 which extract from the volumetric data and store them into HDF5. I was successful to randomly extract the patches from raw data and its label by extract_patch_from_volumetric_data function.

I want to store these patches into the HDF5 data. The bellow code performs the task. Could you look at and verify help me my implementation? I am wondering about initial matrix raw_patches, label_patches (lines 2,3), order matrix storing in lines 8,9,10,11). Thank you for reading

num_sample=1000;
raw_patches = np.zeros([num_sample, 1, 50,50,50], dtype=np.float16)
label_patches = np.zeros([num_sample, 1, 50,50,50], dtype=np.int8)
for i in range(num_sample):
    raw_patch, label_patch = extract_patch_from_volumetric_data(in_data)
    #raw_patch shape: [50x50x50], label_patch shape [50x50x50]
    # Store them in a array
    raw_patches[i, 0, :, :, :] = raw_patch  
    label_patches[i, 0, :, :, :] = label_patch        

raw_patches = raw_patches[0:num_sample, :, :, :, :]
label_patches = label_patches[0:num_sample, :, :, :, :]
#Store in HDF5 and txt path
with h5py.File('./trainMS_%s.h5' % index_file, 'w') as f:
    f['data'] = raw_patches
    f['label'] = label_patches

with open('./trainMS_list.txt', 'a') as f:
    f.write('./trainMS_%s.h5\n' % index_file)

This is my prototxt

layer {
  name: "data"
  type: "HDF5Data"
  top: "data"
  top: "label"
  include {
    phase: TRAIN
  }
  hdf5_data_param {
    source: "./trainMS_list.txt"
    batch_size: 1
  }
}
layer {
  name: "conv1a"
  type: "Convolution"
  bottom: "data"
  top: "conv1a"
  convolution_param {
    num_output: 32
    pad: 1
    kernel_size: 3
    stride: 1
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
    }
  }
}

Update: Script to test convolution

filename='./trainMS_0.h5'
with h5py.File(filename) as hf:
    data=hf.get('data')
    npdata=np.array(data)
    print npdata.shape   # 100x1x50x50x50
    label = hf.get('label')
    nplabel = np.array(label)
    print nplabel.shape  # 100x50x50x50

    caffe.set_mode_cpu()
    net = caffe.Net('conv.prototxt', caffe.TEST)
    im =npdata[20,0,:,:,:] # input is 50x50x50
    im_input = im[np.newaxis, np.newaxis, :, :]
    net.blobs['data'].reshape(*im_input.shape)
    net.blobs['data'].data[...] = im_input

    # pick first filter output
    conv0 = net.blobs['conv'].data[0, 0]
    print("pre-surgery output mean {:.2f}".format(conv0.mean())) #output is 0
    # set first filter bias to 1
    net.params['conv'][1].data[0] = 1.
    net.forward()
    conv_out=net.blobs['conv'].data[0, 0];
    print conv_out.shape # 50x50x50

来源：https://stackoverflow.com/questions/42310081/how-to-store-volumetric-patch-into-hdf5