Vanilla CNN caffe model

readme.md

name	caffemodel	caffemodel_url	license	sha1	caffe_commit
Vanilla CNN Model	vanillaCNN.caffemodel	https://github.com/ishay2b/VanillaCNN/blob/master/ZOO/vanillaCNN.caffemodel?raw=true	unrestricted	b5e34ce75d078025e07452cb47e65d198fe27912	9c9f94e18a8909580a6b94c44dbb1e46f0ee8eb8

Implementation of the Vanilla CNN described in the paper: Yue Wu and Tal Hassner, "Facial Landmark Detection with Tweaked Convolutional Neural Networks", arXiv preprint arXiv:1511.04031, 12 Nov. 2015. See project page for more information about this project. http://www.openu.ac.il/home/hassner/projects/tcnn_landmarks/

Written by Ishay Tubi: https://www.linkedin.com/in/ishay2b

This software is provided as is, without any warranty, with no legal constraints.

See github repository: http://bit.ly/1SbTqTf

The bundled model is the iteration 1,000,000 snapshot using vanilla_adam_solver.prototxt

Model was traind using AFLW train data, and obtains a 7.6% mean error for AFW benchmark and 8.3% for AFLW benchmark.

License

This model is released for unrestricted use.

vanilla_adam_solver.prototxt

# The train/test net protocol buffer definition
# this follows "ADAM: A METHOD FOR STOCHASTIC OPTIMIZATION"
net: "vanilla_train.prototxt"
type: "Adam"
test_iter: 1000
test_interval: 5000
base_lr: 0.00001
momentum: 0.9
momentum2: 0.999
# since Adam dynamically changes the learning rate, we set the base learning
# rate to a fixed value
lr_policy: "fixed"
# Display every 1000 iterations
display: 2500
# The maximum number of iterations
max_iter: 1400000
# snapshot intermediate results
snapshot: 5000
snapshot_prefix: "../caffeData/snapshots/snap"
# solver mode: CPU or GPU
solver_mode: CPU

vanilla_deploy.prototxt

name: "vanila"
input: "data"
input_dim: 1
input_dim: 3
input_dim: 40
input_dim: 40
########################################
# the actual net
# layer 1
layer {
    name: "Conv1"
    type: "Convolution"
    bottom: "data"
    top: "Conv1"
    param {
        lr_mult: 1
        decay_mult: 1
    }
    param {
        lr_mult: 2
        decay_mult: 0
    }
    convolution_param {
        num_output: 16
        pad: 2
        kernel_size: 5
        stride: 1
        weight_filler {
            type: "xavier"
            std: 0.1
        }
        bias_filler {
            type: "constant"
            value: 0.2
        }
    }
}

layer {
  name: "ActivationTangH1"
  bottom: "Conv1"
  top: "ActivationTangH1"
  type: "TanH"
}

layer {
  name: "ActivationAbs1"
  bottom: "ActivationTangH1"
  top: "Abs1"
  type: "AbsVal"
}

layer {
    name: "Pool1"
    type: "Pooling"
    bottom: "Abs1"
    top: "Pool1"
    pooling_param {
        pool: MAX
        kernel_size: 2
        stride: 2
    }
}

layer {
    name: "Conv2"
    type: "Convolution"
    bottom: "Pool1"
    top: "Conv2"
    param {
        lr_mult: 1
        decay_mult: 1
    }
    param {
        lr_mult: 2
        decay_mult: 0
    }
    convolution_param {
        num_output: 48
        pad: 1
        kernel_size: 3
        stride: 1
        weight_filler {
        type: "xavier"
        std: 0.1
    }
    bias_filler {
        type: "constant"
        value: 0.2
        }
    }
}

layer {
  name: "ActivationTangH2"
  bottom: "Conv2"
  top: "ActivationTangH2"
  type: "TanH"
}

layer {
  name: "ActivationAbs2"
  bottom: "ActivationTangH2"
  top: "Abs2"
  type: "AbsVal"
}


layer {
    name: "Pool2"
    type: "Pooling"
    bottom: "Abs2"
    top: "Pool2"
    pooling_param {
        pool: MAX
        kernel_size: 2
        stride: 2
    }
}

# layer 3
layer {
    name: "Conv3"
    type: "Convolution"
    bottom: "Pool2"
    top: "Conv3"
    param {
        lr_mult: 1
        decay_mult: 1
    }
    param {
        lr_mult: 2
        decay_mult: 0
    }
    convolution_param {
        num_output: 64
        pad: 0
        kernel_size: 3
        stride: 1
        weight_filler {
            type: "xavier"
            std: 0.1
        }
        bias_filler {
            type: "constant"
            value: 0.2
        }
    }
}


layer {
  name: "ActivationTangH3"
  bottom: "Conv3"
  top: "ActivationTangH3"
  type: "TanH"
}

layer {
  name: "ActivationAbs3"
  bottom: "ActivationTangH3"
  top: "Abs3"
  type: "AbsVal"
}

layer {
    name: "Pool3"
    type: "Pooling"
    bottom: "Abs3"
    top: "Pool3"
    pooling_param {
        pool: MAX
        kernel_size: 3
        stride: 2
    }
}

# layer 4
layer {
    name: "Conv4"
    type: "Convolution"
    bottom: "Pool3"
    top: "Conv4"
    param {
        lr_mult: 1
        decay_mult: 1
    }
    param {
        lr_mult: 2
        decay_mult: 0
    }
    convolution_param {
        num_output: 64
        pad: 0
        kernel_size: 2
        stride: 1
        weight_filler {
            type: "xavier"
            std: 0.1
        }
        bias_filler {
            type: "constant"
            value: 0.2
        }
    }
}


layer {
  name: "ActivationTangH4"
  bottom: "Conv4"
  top: "ActivationTangH4"
  type: "TanH"
}

layer {
  name: "ActivationAbs4"
  bottom: "ActivationTangH4"
  top: "Abs4"
  type: "AbsVal"
}


########################################

layer {
    name: "Dense1"
    type: "InnerProduct"
    bottom: "Abs4"
    top: "Dense1"
    param {
        lr_mult: 1
        decay_mult: 1
    }
    param {
        lr_mult: 2
        decay_mult: 0
    }
    inner_product_param {
        num_output: 100
        weight_filler {
            type: "xavier"
        }
        bias_filler {
            type: "constant"
            value: 0
        }
    }
}


layer {
  name: "ActivationTangH5"
  bottom: "Dense1"
  top: "ActivationTangH5"
  type: "TanH"
}

layer {
  name: "ActivationAbs5"
  bottom: "ActivationTangH5"
  top: "Abs5"
  type: "AbsVal"
}


layer {
    name: "Dense2"
    type: "InnerProduct"
    bottom: "Abs5"
    top: "Dense2"
    param {
        lr_mult: 1
        decay_mult: 1
    }
    param {
        lr_mult: 2
        decay_mult: 0
    }
    inner_product_param {
        num_output: 10
        weight_filler {
            type: "xavier"
        }
        bias_filler {
            type: "constant"
            value: 0
        }
    }
}

vanilla_train.prototxt

name: "vanila"

layer {
   type: "HDF5Data"
   top: "X" # same name as given in create_dataset!
   top: "landmarks"
   hdf5_data_param {
     source: "/Users/ishay/Dev/VanilaCNN/caffeData/train.txt" # do not give the h5 files directly, but the list.
     batch_size: 30
   }
   include { phase:TRAIN }
 }

layer {
   type: "HDF5Data"
   top: "X" # same name as given in create_dataset!
   top: "landmarks"
   hdf5_data_param {
     source: "/Users/ishay/Dev/VanilaCNN/caffeData/test.txt" # do not give the h5 files directly, but the list.
     batch_size: 10
   }
   include { phase:TEST }
}

layer {
    name: "Conv1"
    type: "Convolution"
    bottom: "X"
    top: "Conv1"
    param {
        lr_mult: 1
        decay_mult: 1
    }
    param {
        lr_mult: 2
        decay_mult: 0
    }
    convolution_param {
        num_output: 16
        pad: 2
        kernel_size: 5
        stride: 1
        weight_filler {
            type: "xavier"
            std: 0.1
        }
        bias_filler {
            type: "constant"
            value: 0.2
        }
    }
}

layer {
  name: "ActivationTangH1"
  bottom: "Conv1"
  top: "ActivationTangH1"
  type: "TanH"
}

layer {
  name: "ActivationAbs1"
  bottom: "ActivationTangH1"
  top: "Abs1"
  type: "AbsVal"
}

layer {
    name: "Pool1"
    type: "Pooling"
    bottom: "Abs1"
    top: "Pool1"
    pooling_param {
        pool: MAX
        kernel_size: 2
        stride: 2
    }
}

layer {
    name: "Conv2"
    type: "Convolution"
    bottom: "Pool1"
    top: "Conv2"
    param {
        lr_mult: 1
        decay_mult: 1
    }
    param {
        lr_mult: 2
        decay_mult: 0
    }
    convolution_param {
        num_output: 48
        pad: 1
        kernel_size: 3
        stride: 1
        weight_filler {
        type: "xavier"
        std: 0.1
    }
    bias_filler {
        type: "constant"
        value: 0.2
        }
    }
}

layer {
  name: "ActivationTangH2"
  bottom: "Conv2"
  top: "ActivationTangH2"
  type: "TanH"
}

layer {
  name: "ActivationAbs2"
  bottom: "ActivationTangH2"
  top: "Abs2"
  type: "AbsVal"
}


layer {
    name: "Pool2"
    type: "Pooling"
    bottom: "Abs2"
    top: "Pool2"
    pooling_param {
        pool: MAX
        kernel_size: 2
        stride: 2
    }
}

# layer 3
layer {
    name: "Conv3"
    type: "Convolution"
    bottom: "Pool2"
    top: "Conv3"
    param {
        lr_mult: 1
        decay_mult: 1
    }
    param {
        lr_mult: 2
        decay_mult: 0
    }
    convolution_param {
        num_output: 64
        pad: 0
        kernel_size: 3
        stride: 1
        weight_filler {
            type: "xavier"
            std: 0.1
        }
        bias_filler {
            type: "constant"
            value: 0.2
        }
    }
}


layer {
  name: "ActivationTangH3"
  bottom: "Conv3"
  top: "ActivationTangH3"
  type: "TanH"
}

layer {
  name: "ActivationAbs3"
  bottom: "ActivationTangH3"
  top: "Abs3"
  type: "AbsVal"
}

layer {
    name: "Pool3"
    type: "Pooling"
    bottom: "Abs3"
    top: "Pool3"
    pooling_param {
        pool: MAX
        kernel_size: 3
        stride: 2
    }
}

# layer 4
layer {
    name: "Conv4"
    type: "Convolution"
    bottom: "Pool3"
    top: "Conv4"
    param {
        lr_mult: 1
        decay_mult: 1
    }
    param {
        lr_mult: 2
        decay_mult: 0
    }
    convolution_param {
        num_output: 64
        pad: 0
        kernel_size: 2
        stride: 1
        weight_filler {
            type: "xavier"
            std: 0.1
        }
        bias_filler {
            type: "constant"
            value: 0.2
        }
    }
}


layer {
  name: "ActivationTangH4"
  bottom: "Conv4"
  top: "ActivationTangH4"
  type: "TanH"
}

layer {
  name: "ActivationAbs4"
  bottom: "ActivationTangH4"
  top: "Abs4"
  type: "AbsVal"
}


########################################

layer {
    name: "Dense1"
    type: "InnerProduct"
    bottom: "Abs4"
    top: "Dense1"
    param {
        lr_mult: 1
        decay_mult: 1
    }
    param {
        lr_mult: 2
        decay_mult: 0
    }
    inner_product_param {
        num_output: 100
        weight_filler {
            type: "xavier"
        }
        bias_filler {
            type: "constant"
            value: 0
        }
    }
}


layer {
  name: "ActivationTangH5"
  bottom: "Dense1"
  top: "ActivationTangH5"
  type: "TanH"
}

layer {
  name: "ActivationAbs5"
  bottom: "ActivationTangH5"
  top: "Abs5"
  type: "AbsVal"
}


layer {
    name: "Dense2"
    type: "InnerProduct"
    bottom: "Abs5"
    top: "Dense2"
    param {
        lr_mult: 1
        decay_mult: 1
    }
    param {
        lr_mult: 2
        decay_mult: 0
    }
    inner_product_param {
        num_output: 10
        weight_filler {
            type: "xavier"
        }
        bias_filler {
            type: "constant"
            value: 0
        }
    }
}

layer {
  type: 'Python'
  name: 'loss'
  top: 'loss'
  bottom: 'Dense2'
  bottom: 'landmarks'
  python_param {
    # the module name -- usually the filename -- that needs to be in $PYTHONPATH
    module: 'python.NormlizedMSE'
    # the layer name -- the class name in the module
    layer: 'NormlizedMSE'
  }
  # set loss weight so Caffe knows this is a loss layer.
  # since PythonLayer inherits directly from Layer, this isn't automatically
  # known to Caffe
  loss_weight: 1
}

Hi,

I tried to get the model with ./scripts/download_model_from_gist.sh <gist_id>, but the result is a html file instead, not a zip file.

Can you please fix this problem? although I can get the model from your github project.

In your paper, it's mentioned TCDCN can be used to detect more than 5 Landmark points. But i can't find trained model weights or deploy.prototxt for that purpose. Can you please redirect me to that.