DeepDrive - Vision model that drives in GTAV

readme.md

name	DeepDrive
caffemodel	caffe_deep_drive_train_iter_35352.caffemodel
caffemodel_url	https://docs.google.com/uc?export=download&id=0B2UgaM91sqeAN0dJb05jWThJUUk
license	MIT
sha1	448c5fe310a8ac954f2dbb7d6e19733edad63c6f
caffe_commit	843575eca2582fc30588ceaf31c987ad7707033d
gist_id	66e69c10c9ec82b299279bc7609544d2

This model stays inside the lane and stops for traffic in various weather and lighting conditions in GTAV when viewing images streamed from a hood-mounted camera. It was trained on top of BVLC's reference AlexNet model. See deepdrive.io for complete details.

License

MIT

deep_drive_model.prototxt

name: "GTANet"

# Can be used with pretrained Caffenet (AlexNet architecture).
# Layers with names containing 'gtanet' are not transferred from Caffenet.

layer {
  name: "gta_frames_input_layer"
  type: "HDF5Data"
  top: "images"
  top: "targets"
  hdf5_data_param {
    source: "data/deep_drive/full_train_h5_list.txt"
    batch_size: 64
    shuffle: true
  }
  include { phase:TRAIN }
}

layer {
  name: "gta_frames_input_layer"
  type: "HDF5Data"
  top: "images"
  top: "targets"
  hdf5_data_param {
    source: "data/deep_drive/full_test_h5_list.txt"
    batch_size: 64
  }
  include { phase:TEST }
}

layer {
  name: "conv1"
  type: "Convolution"
  bottom: "images"
  top: "conv1"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 96
    kernel_size: 11
    stride: 4
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
	bias_filler {
	  type: "constant"
	  value: 0
	}
  }
}

layer {
  name: "relu1"
  type: "ReLU"
  bottom: "conv1"
  top: "conv1"
}

layer {
  name: "pool1"
  type: "Pooling"
  bottom: "conv1"
  top: "pool1"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}

layer {
  name: "norm1"
  type: "LRN"
  bottom: "pool1"
  top: "norm1"
  lrn_param {
    local_size: 5
    alpha: 0.0001
    beta: 0.75
  }
}

layer {
  name: "conv2"
  type: "Convolution"
  bottom: "norm1"
  top: "conv2"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    pad: 2
    kernel_size: 5
    group: 2
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}

layer {
  name: "relu2"
  type: "ReLU"
  bottom: "conv2"
  top: "conv2"
}

layer {
  name: "pool2"
  type: "Pooling"
  bottom: "conv2"
  top: "pool2"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}

layer {
  name: "norm2"
  type: "LRN"
  bottom: "pool2"
  top: "norm2"
  lrn_param {
    local_size: 5
    alpha: 0.0001
    beta: 0.75
  }
}

layer {
  name: "conv3"
  type: "Convolution"
  bottom: "norm2"
  top: "conv3"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 384
    pad: 1
    kernel_size: 3
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu3"
  type: "ReLU"
  bottom: "conv3"
  top: "conv3"
}

layer {
  name: "conv4"
  type: "Convolution"
  bottom: "conv3"
  top: "conv4"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
   convolution_param {
    num_output: 384
    pad: 1
    kernel_size: 3
    group: 2
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}

layer {
  name: "relu4"
  type: "ReLU"
  bottom: "conv4"
  top: "conv4"
}

layer {
  name: "conv5"
  type: "Convolution"
  bottom: "conv4"
  top: "conv5"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    pad: 1
    kernel_size: 3
    group: 2
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}

layer {
  name: "relu5"
  type: "ReLU"
  bottom: "conv5"
  top: "conv5"
}

layer {
  name: "pool5"
  type: "Pooling"
  bottom: "conv5"
  top: "pool5"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}

layer {
  name: "fc6_gtanet"
  type: "InnerProduct"
  bottom: "pool5"
  top: "fc6"
  param {
    lr_mult: 1     # learning rate multiplier for the filters
    decay_mult: 1  # weight decay multiplier for the filters
  }
  param {
    lr_mult: 2     # learning rate multiplier for the biases
    decay_mult: 0  # weight decay multiplier for the biases
  }
  inner_product_param {
    num_output: 4096
    weight_filler {
      type: "gaussian"
      std: 0.005
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}

layer {
  name: "relu6_gtanet"
  type: "ReLU"
  bottom: "fc6"
  top: "fc6"
}

layer {
  name: "drop6_gtanet"
  type: "Dropout"
  bottom: "fc6"
  top: "fc6"
  dropout_param {
    dropout_ratio: 0.5
  }
}


layer {
  name: "fc7_gtanet"
  type: "InnerProduct"
  bottom: "fc6"
  top: "fc7"
  param {
    lr_mult: 1    # learning rate multiplier for the filters
    decay_mult: 1 # weight decay multiplier for the filters
  }
  param {
    lr_mult: 2    # learning rate multiplier for the biases
    decay_mult: 0 # weight decay multiplier for the biases
  }
  inner_product_param {
    num_output: 4096
    weight_filler {
      type: "gaussian"
      std: 0.005
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}

layer {
  name: "relu7_gtanet"
  type: "ReLU"
  bottom: "fc7"
  top: "fc7"
}

layer {
  name: "drop7_gtanet"
  type: "Dropout"
  bottom: "fc7"
  top: "fc7"
  dropout_param {
    dropout_ratio: 0.05
  }
}

# The output layer is a fully-connected linear layer with a single output for each valid action.
layer {
  name: "gtanet_fctop"
  type: "InnerProduct"
  bottom: "fc7"
  top: "gtanet_fctop"
  param {
    lr_mult: 1    # learning rate multiplier for the filters
    decay_mult: 1 # weight decay multiplier for the filters
  }
  param {
    lr_mult: 2    # learning rate multiplier for the biases
    decay_mult: 0 # weight decay multiplier for the biases
  }
  inner_product_param {
    num_output: 6 # Number of output neurons
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "loss"
  type: "EuclideanLoss"
  bottom: "gtanet_fctop"
  bottom: "targets"
  top: "loss"
}

deep_drive_solver.prototxt

net: "examples/deep_drive/deep_drive_model.prototxt"
test_iter: 1000
test_interval: 2000
type: "Adam"
momentum: 0.8
momentum2: 0.9
base_lr: 0.0002
# manually decay to the following when learning subsides
#base_lr: 0.00002
#base_lr: 0.000002
#base_lr: 0.0000002
#base_lr: 0.00000002
#base_lr: 0.000000002
lr_policy: "step" # learning rate policy: drop the learning rate in "steps"
                  # by a factor of gamma every stepsize iterations
gamma: 0.1        # drop the learning rate by this factor
stepsize: 20000   # drop the learning rate by above every stepsize iterations
snapshot: 20000
snapshot_prefix: "caffe_deep_drive_train"
max_iter: 50000000
weight_decay: 0.0005
display: 40
solver_mode: GPU
test_initialization: false
#debug_info: true

super cool!!

If you don't mind me asking, where can I get the deepdrive.tfweights?