train_val.prototxt - Using data_param and image_data_param together

[Sid]

My train_val.prototxt:

layer {
  name: "data"
  type: "Data"
  top: "data"
  top: "label"
  include {
    phase: TRAIN
  }
  transform_param {
    mirror: true
    crop_size: 224
    mean_file: "data/mean.binaryproto"
  }
  data_param {
    source: "data/train_lmdb"
    batch_size: 32
    backend: LMDB
  }
  image_data_param {
    new_height: 256
    new_width: 256
  }
}

My Network is running but I want to make sure that this layer definition will work properly. How can i ensure that the images being fed to the network during training will get resized to 256x256 and then cropped to 224? The reason I can't just use image_data_param is because I can't use the backend keyword within image_data_param (error is thrown)

Any help would be greatly appreciated.

Thanks!

[Sid]

I looked at my solver log and there is no output from image_data_layer.cpp. line 80 of image_data_layer.cpp indicates that there should be an output to the log but in my case, this did not occur. So, it looks like my image_data_param configuration is being ignored. The only data layer log output returned was the following:

I1231 21:35:02.604929 25749 data_layer.cpp:44] output data size: 26,3,224,224

So it looks like the cropping is happening but the initial resize that I want is not. Does anyone know of a way to first resize and then crop?