How to create an caffemodel file from training image and its labeled?

Question

I am working in age classification based on the opensource at here The python code has

age_net_pretrained='./age_net.caffemodel'
age_net_model_file='./deploy_age.prototxt'
age_net = caffe.Classifier(age_net_model_file, age_net_pretrained,
       channel_swap=(2,1,0),
       raw_scale=255,
       image_dims=(256, 256))

In which .prototxt file is shown as below. I remain one file that is ".caffemodel". As the source code, he provided it before. However, I would like to create it again based on my face database. Could you have any tutorial or some way to create it? I assume that I have a folder image that include 100 images and divided belongs to each age groups (1 to 1) such as

image1.png 1
image2.png 1
..
image10.png 1
image11.png 2
image12.png 2
...
image100.png 10

This is prototxt file. Thanks in advance

name: "CaffeNet"
input: "data"
input_dim: 1
input_dim: 3
input_dim: 227
input_dim: 227
layers {
  name: "conv1"
  type: CONVOLUTION
  bottom: "data"
  top: "conv1"
  convolution_param {
    num_output: 96
    kernel_size: 7
    stride: 4
  }
}
layers {
  name: "relu1"
  type: RELU
  bottom: "conv1"
  top: "conv1"
}
layers {
  name: "pool1"
  type: POOLING
  bottom: "conv1"
  top: "pool1"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layers {
  name: "norm1"
  type: LRN
  bottom: "pool1"
  top: "norm1"
  lrn_param {
    local_size: 5
    alpha: 0.0001
    beta: 0.75
  }
}
layers {
  name: "conv2"
  type: CONVOLUTION
  bottom: "norm1"
  top: "conv2"
  convolution_param {
    num_output: 256
    pad: 2
    kernel_size: 5
  }
}
layers {
  name: "relu2"
  type: RELU
  bottom: "conv2"
  top: "conv2"
}
layers {
  name: "pool2"
  type: POOLING
  bottom: "conv2"
  top: "pool2"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layers {
  name: "norm2"
  type: LRN
  bottom: "pool2"
  top: "norm2"
  lrn_param {
    local_size: 5
    alpha: 0.0001
    beta: 0.75
  }
}
layers {
  name: "conv3"
  type: CONVOLUTION
  bottom: "norm2"
  top: "conv3"
  convolution_param {
    num_output: 384
    pad: 1
    kernel_size: 3
  }
}
layers{
  name: "relu3" 
  type: RELU
  bottom: "conv3"
  top: "conv3"
}
layers {
  name: "pool5"
  type: POOLING
  bottom: "conv3"
  top: "pool5"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layers {
  name: "fc6"
  type: INNER_PRODUCT
  bottom: "pool5"
  top: "fc6"
  inner_product_param {
    num_output: 512
  }
}
layers {
  name: "relu6"
  type: RELU
  bottom: "fc6"
  top: "fc6"
}
layers {
  name: "drop6"
  type: DROPOUT
  bottom: "fc6"
  top: "fc6"
  dropout_param {
    dropout_ratio: 0.5
  }
}
layers {
  name: "fc7"
  type: INNER_PRODUCT
  bottom: "fc6"
  top: "fc7"
  inner_product_param {
    num_output: 512
  }
}
layers {
  name: "relu7"
  type: RELU
  bottom: "fc7"
  top: "fc7"
}
layers {
  name: "drop7"
  type: DROPOUT
  bottom: "fc7"
  top: "fc7"
  dropout_param {
    dropout_ratio: 0.5
  }
}
layers {
  name: "fc8"
  type: INNER_PRODUCT
  bottom: "fc7"
  top: "fc8"
  inner_product_param {
    num_output: 8
  }
}
layers {
  name: "prob"
  type: SOFTMAX
  bottom: "fc8"
  top: "prob"
}

Solution

To get a caffemodel you need to train the network. That prototxt file is only to deploy the model and cannot be used to train it.

You need to add a data layer that points to your database. To use a list of files as you mention, the source of the layer should be HDF5. You will probably want to add a transform_param with the mean value. The image files can be replaced by a LMDB or LevelDB database for efficiency purposes.

At the end of the network you will have to substitute the 'prob' layer with a 'loss' layer. Something like this:

layers { name: "loss" type: SoftmaxWithLoss bottom: "fc8" top: "loss" }

The layer catalogue can be found here:

http://caffe.berkeleyvision.org/tutorial/layers.html

Or, as your network is a well known one... just look at this tutorial :P.

http://caffe.berkeleyvision.org/gathered/examples/imagenet.html

The correct prototxt file for training is included in caffe ('train_val.prototxt').

Comments

[omatai]

You say that this prototxt file is only to deploy the model. Is that because the first input_dim specified is 1, and to train you need to specify a reasonable batch size?

[MacKa]

Hi, I'm new to this area. I would like to create my own .caffemodel and convert it to CoreML. How would I do that? Please guide me.