Can I create a data source for a caffe model from java?

[altus88]

I am trying to run training a caffe model from java. I need to generate a lmdb database on the fly. There is some python code http://deepdish.io/2015/04/28/creating-lmdb-in-python/ which can do it, but in java I can not do the same since Datum class does not have method "SerializeToString()". Another approach I tried is using "MemoryData" data layer, and feed training data directly to the model:

SolverParameter solver_param = new SolverParameter();
ReadProtoFromTextFileOrDie(path/to/solver, solver_param); 
FloatSolver solver = FloatSolverRegistry.CreateSolver(solver_param);
FloatMemoryDataLayer input =  solver.net().layer_by_name(FloatMemoryDataLayer.class, 'mnist');

MatVector matVector = new MatVector();
IntPointer intPoiter = new IntPointer();

Mat img = opencv_imgcodecs.imread(path_to_img, opencv_imgcodecs.IMREAD_GRAYSCALE);
Mat img_res_float = new Mat();
img.convertTo(img_res_float, opencv_core.CV_32FC1);

matVector.put(img_res_float);
intPoiter.put(new Integer(0));

input.AddMatVector(matVector, intPoiter);
solver.Solve(new BytePointer());

Prototxt of the model:

name: "LeNet"
layer {
  name: "mnist"
  type: "MemoryData"
  top: "data"
  top: "label"
  include {
    phase: TRAIN
  }
  transform_param {
    scale: 0.00390625
  }
  memory_data_param {
    batch_size: 1
    channels: 1
    height: 28
    width: 28
  }
}
layer {
  name: "conv1"
  type: "Convolution"
  bottom: "data"
  top: "conv1"
  param {
    lr_mult: 1
  }
  param {
    lr_mult: 2
  }
  convolution_param {
    num_output: 20
    kernel_size: 5
    stride: 1
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
    }
  }
}
layer {
  name: "pool1"
  type: "Pooling"
  bottom: "conv1"
  top: "pool1"
  pooling_param {
    pool: MAX
    kernel_size: 2
    stride: 2
  }
}
layer {
  name: "conv2"
  type: "Convolution"
  bottom: "pool1"
  top: "conv2"
  param {
    lr_mult: 1
  }
  param {
    lr_mult: 2
  }
  convolution_param {
    num_output: 50
    kernel_size: 5
    stride: 1
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
    }
  }
}
layer {
  name: "pool2"
  type: "Pooling"
  bottom: "conv2"
  top: "pool2"
  pooling_param {
    pool: MAX
    kernel_size: 2
    stride: 2
  }
}
layer {
  name: "ip1"
  type: "InnerProduct"
  bottom: "pool2"
  top: "ip1"
  param {
    lr_mult: 1
  }
  param {
    lr_mult: 2
  }
  inner_product_param {
    num_output: 500
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
    }
  }
}
layer {
  name: "relu1"
  type: "ReLU"
  bottom: "ip1"
  top: "ip1"
}
layer {
  name: "ip2"
  type: "InnerProduct"
  bottom: "ip1"
  top: "ip2"
  param {
    lr_mult: 1
  }
  param {
    lr_mult: 2
  }
  inner_product_param {
    num_output: 10
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
    }
  }
}
layer {
  name: "loss"
  type: "SoftmaxWithLoss"
  bottom: "ip2"
  bottom: "label"
  top: "loss"
}

but at line
intPoiter.put(new Integer(0));
I get an error "Exception in thread "main" java.lang.NullPointerException: This pointer address is NULL." Could you please suggest something.

[altus88]

If I write
input1.AddMatVector(matVector, new IntPointer(1));

I get the following error:
F0407 11:34:50.119905 6363 blob.cpp:32] Check failed: shape[i] >= 0 (-1527131614 vs. 0)
It seems something wrong with the shape of the mat, but I checked, it is ok: (1,28,28)

[saudet]

I've added the API that contains SerializeToString(), so if that works, please try to use that! Thanks

[altus88]

Thanks! I will try and let you know

[saudet]

Included in version 1.2. Enjoy!