mobilenet.export.yaml

---
dataset:
    background_class: {has: true, use: true}
    channel_means: [0.485019607843137, 0.4579607843, 0.407607843137255]
    name: imagenet
    num_classes: 1001
    num_examples_per_epoch: {train: 1281167, validate: 50000}
    path: {train: imagenet/train-*, validate: imagenet/validation-*}
    shape: {channels: 3, height: 224, width: 224}
depth_multiplier: 1
hyper:
- &id002 {scale: 4.0e-05, type: tensorflow.contrib.layers.l2_regularizer}
-   kernel_size: [3, 3]
    normalizer_fn: &id001 {center: true, decay: 0.9997, epsilon: 0.001,
        scale: true, type: tensorflow.contrib.slim.batch_norm}
    padding: same
    type: convolution
    weights_initializer: &id003 {stddev: 0.09, type: tensorflow.truncated_normal_initializer}
-   depth_multiplier: 1
    depthwise_regularizer: null
    kernel_size: [3, 3]
    normalizer_fn: *id001
    padding: same
    pointwise_regularizer: *id002
    type: depthwise_separable_convolution
    weights_initializer: *id003
-   kernel_size: [7, 7]
    padding: valid
    stride: 2
    type: average_pool
- {keep_prob: 0.999, type: dropout}
-   activation_fn: null
    kernel_size: [1, 1]
    normalizer_fn: null
    num_outputs: num_classes
    type: convolution
    weights_regularizer: *id002
logits: logits
name: mobilenet_v1
net:
-   kernel_size: [3, 3]
    name: conv0
    normalizer_fn: *id001
    num_outputs: 32
    padding: same
    stride: 2
    type: convolution
    weights_initializer: *id003
    weights_regularizer: *id002
-   depth_multiplier: 1
    depthwise_regularizer: null
    kernel_size: [3, 3]
    name: conv1
    normalizer_fn: *id001
    num_outputs: 64
    padding: same
    pointwise_regularizer: *id002
    stride: 1
    type: depthwise_separable_convolution
    weights_initializer: *id003
-   depth_multiplier: 1
    depthwise_regularizer: null
    kernel_size: [3, 3]
    name: conv2
    normalizer_fn: *id001
    num_outputs: 128
    padding: same
    pointwise_regularizer: *id002
    stride: 2
    type: depthwise_separable_convolution
    weights_initializer: *id003
-   depth_multiplier: 1
    depthwise_regularizer: null
    kernel_size: [3, 3]
    name: conv3
    normalizer_fn: *id001
    num_outputs: 128
    padding: same
    pointwise_regularizer: *id002
    stride: 1
    type: depthwise_separable_convolution
    weights_initializer: *id003
-   depth_multiplier: 1
    depthwise_regularizer: null
    kernel_size: [3, 3]
    name: conv4
    normalizer_fn: *id001
    num_outputs: 256
    padding: same
    pointwise_regularizer: *id002
    stride: 2
    type: depthwise_separable_convolution
    weights_initializer: *id003
-   depth_multiplier: 1
    depthwise_regularizer: null
    kernel_size: [3, 3]
    name: conv5
    normalizer_fn: *id001
    num_outputs: 256
    padding: same
    pointwise_regularizer: *id002
    stride: 1
    type: depthwise_separable_convolution
    weights_initializer: *id003
-   depth_multiplier: 1
    depthwise_regularizer: null
    kernel_size: [3, 3]
    name: conv6
    normalizer_fn: *id001
    num_outputs: 512
    padding: same
    pointwise_regularizer: *id002
    stride: 2
    type: depthwise_separable_convolution
    weights_initializer: *id003
-   depth_multiplier: 1
    depthwise_regularizer: null
    kernel_size: [3, 3]
    name: conv7
    normalizer_fn: *id001
    num_outputs: 512
    padding: same
    pointwise_regularizer: *id002
    stride: 1
    type: depthwise_separable_convolution
    weights_initializer: *id003
-   depth_multiplier: 1
    depthwise_regularizer: null
    kernel_size: [3, 3]
    name: conv8
    normalizer_fn: *id001
    num_outputs: 512
    padding: same
    pointwise_regularizer: *id002
    stride: 1
    type: depthwise_separable_convolution
    weights_initializer: *id003
-   depth_multiplier: 1
    depthwise_regularizer: null
    kernel_size: [3, 3]
    name: conv9
    normalizer_fn: *id001
    num_outputs: 512
    padding: same
    pointwise_regularizer: *id002
    stride: 1
    type: depthwise_separable_convolution
    weights_initializer: *id003
-   depth_multiplier: 1
    depthwise_regularizer: null
    kernel_size: [3, 3]
    name: conv10
    normalizer_fn: *id001
    num_outputs: 512
    padding: same
    pointwise_regularizer: *id002
    stride: 1
    type: depthwise_separable_convolution
    weights_initializer: *id003
-   depth_multiplier: 1
    depthwise_regularizer: null
    kernel_size: [3, 3]
    name: conv11
    normalizer_fn: *id001
    num_outputs: 512
    padding: same
    pointwise_regularizer: *id002
    stride: 1
    type: depthwise_separable_convolution
    weights_initializer: *id003
-   depth_multiplier: 1
    depthwise_regularizer: null
    kernel_size: [3, 3]
    name: conv12
    normalizer_fn: *id001
    num_outputs: 1024
    padding: same
    pointwise_regularizer: *id002
    stride: 2
    type: depthwise_separable_convolution
    weights_initializer: *id003
-   depth_multiplier: 1
    depthwise_regularizer: null
    kernel_size: [3, 3]
    name: conv13
    normalizer_fn: *id001
    num_outputs: 1024
    padding: same
    pointwise_regularizer: *id002
    stride: 1
    type: depthwise_separable_convolution
    weights_initializer: *id003
-   kernel_size: [7, 7]
    name: pool
    padding: valid
    stride: 2
    type: average_pool
- {keep_prob: 0.999, name: dropout, type: dropout}
-   activation_fn: null
    kernel_size: [1, 1]
    name: fc
    normalizer_fn: null
    num_outputs: num_classes
    type: convolution
    weights_regularizer: *id002
-   axis: [1, 2]
    name: logits
    type: squeeze
preprocess:
    final:
    - {scale: 2.0, shift: -1.0, type: linear_map}
    train:
    - {type: distort_bbox}
    - {type: distort_color}
    - {type: random_flip}
    validate:
    - {central_fraction: 0.875, type: central_crop}
system:
    batch_size: 256
    checkpoint: {load: latest, save: true}
    log_level: {mayo: info, tensorflow: 2}
    max_steps: 10000000
    num_gpus: 4
    num_preprocess_threads: 16
    search_paths:
        checkpoints: [checkpoints]
        datasets: [datasets]
        summaries: [summaries]
    use_pdb: true
train:
    learning_rate: {_num_epochs_per_decay: 30.0, decay_rate: 0.16, decay_steps: 150136.7578125,
        learning_rate: 0.1, staircase: true, type: tensorflow.train.exponential_decay}
    moving_average_decay: 0.9999
    optimizer: {decay: 0.9, epsilon: 1.0, momentum: 0.9, type: tensorflow.train.RMSPropOptimizer}