-
Notifications
You must be signed in to change notification settings - Fork 56
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
update MobileFaceNet using static layer name and add training result
- Loading branch information
1 parent
2f1e332
commit 2dae8e6
Showing
11 changed files
with
107 additions
and
137 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -1,123 +1,92 @@ | ||
| from tensorflow.keras import backend as K | ||
| from tensorflow.keras.layers import ( | ||
| Input, | ||
| Conv2D, | ||
| BatchNormalization, | ||
| Layer, | ||
| PReLU, | ||
| SeparableConv2D, | ||
| DepthwiseConv2D, | ||
| add, | ||
| Flatten, | ||
| Dense, | ||
| Dropout, | ||
| GlobalAveragePooling2D, | ||
| Reshape, | ||
| Multiply, | ||
| ) | ||
| from tensorflow.keras.models import Model | ||
|
|
||
| """Building Block Functions""" | ||
|
|
||
|
|
||
| def se_block(inputs, reduction=16): | ||
| channel_axis = 1 if K.image_data_format() == "channels_first" else -1 | ||
| filters = inputs.shape[channel_axis] | ||
| nn = GlobalAveragePooling2D()(inputs) | ||
| nn = Reshape((1, 1, filters))(nn) | ||
| nn = Conv2D(filters // reduction, kernel_size=1)(nn) | ||
| nn = PReLU(shared_axes=[1, 2])(nn) | ||
| nn = Conv2D(filters, kernel_size=1, activation="sigmoid")(nn) | ||
| nn = Multiply()([inputs, nn]) | ||
| import tensorflow as tf | ||
| from tensorflow.keras import layers, models, initializers | ||
|
|
||
|
|
||
| def se_block(inputs, reduction=16, name=""): | ||
| input_channels = inputs.shape[-1] | ||
| nn = layers.GlobalAveragePooling2D(keepdims=True)(inputs) | ||
| # nn = Reshape((1, 1, input_channels))(nn) | ||
| nn = layers.Conv2D(input_channels // reduction, kernel_size=1, name=name + "1_conv")(nn) | ||
| nn = layers.PReLU(shared_axes=[1, 2], alpha_initializer=initializers.Constant(0.25), name=name + "prelu")(nn) | ||
| nn = layers.Conv2D(input_channels, kernel_size=1, name=name + "2_conv")(nn) | ||
| nn = layers.Activation(activation="sigmoid")(nn) | ||
| nn = layers.Multiply(name=name + "out")([inputs, nn]) | ||
| return nn | ||
|
|
||
|
|
||
| def se_block_2(inputs, reduction=16): | ||
| channel_axis = 1 if K.image_data_format() == "channels_first" else -1 | ||
| filters = inputs.shape[channel_axis] | ||
| se = GlobalAveragePooling2D()(inputs) | ||
| se = Dense(filters // reduction, activation="PReLU", use_bias=False)(se) | ||
| se = Dense(filters, activation="sigmoid", use_bias=False)(se) | ||
| # if K.image_data_format() == 'channels_first': | ||
| # se = Permute((3, 1, 2))(se) | ||
| x = Multiply()([inputs, se]) | ||
| return x | ||
|
|
||
|
|
||
| def conv_block(inputs, filters, kernel_size, strides, padding): | ||
| channel_axis = 1 if K.image_data_format() == "channels_first" else -1 | ||
| Z = Conv2D(filters, kernel_size, strides=strides, padding=padding, use_bias=False)(inputs) | ||
| Z = BatchNormalization(axis=channel_axis)(Z) | ||
| A = PReLU(shared_axes=[1, 2])(Z) | ||
| return A | ||
|
|
||
|
|
||
| def separable_conv_block(inputs, filters, kernel_size, strides): | ||
| channel_axis = 1 if K.image_data_format() == "channels_first" else -1 | ||
| Z = SeparableConv2D(filters, kernel_size, strides=strides, padding="same", use_bias=False)(inputs) | ||
| Z = BatchNormalization(axis=channel_axis)(Z) | ||
| A = PReLU(shared_axes=[1, 2])(Z) | ||
| return A | ||
|
|
||
|
|
||
| def bottleneck(inputs, filters, kernel, t, s, r=False, se=False): | ||
| channel_axis = 1 if K.image_data_format() == "channels_first" else -1 | ||
| tchannel = K.int_shape(inputs)[channel_axis] * t | ||
| Z1 = conv_block(inputs, tchannel, 1, 1, "same") | ||
| Z1 = DepthwiseConv2D(kernel, strides=s, padding="same", depth_multiplier=1, use_bias=False)(Z1) | ||
| Z1 = BatchNormalization(axis=channel_axis)(Z1) | ||
| A1 = PReLU(shared_axes=[1, 2])(Z1) | ||
| Z2 = Conv2D(filters, 1, strides=1, padding="same", use_bias=False)(A1) | ||
| Z2 = BatchNormalization(axis=channel_axis)(Z2) | ||
| if se: | ||
| Z2 = se_block(Z2) | ||
| if r: | ||
| Z2 = add([Z2, inputs]) | ||
| return Z2 | ||
|
|
||
|
|
||
| def inverted_residual_block(inputs, filters, kernel, t, strides, n, se=False): | ||
| Z = bottleneck(inputs, filters, kernel, t, strides, se=se) | ||
| for i in range(1, n): | ||
| Z = bottleneck(Z, filters, kernel, t, 1, True, se=se) | ||
| return Z | ||
|
|
||
|
|
||
| def linear_GD_conv_block(inputs, kernel_size, strides): | ||
| channel_axis = 1 if K.image_data_format() == "channels_first" else -1 | ||
| Z = DepthwiseConv2D(kernel_size, strides=strides, padding="valid", depth_multiplier=1, use_bias=False)(inputs) | ||
| Z = BatchNormalization(axis=channel_axis)(Z) | ||
| return Z | ||
|
|
||
|
|
||
| def mobile_facenet(emb_shape=128, input_shape=(112, 112, 3), dropout=1, name="mobile_facenet", weight_file=None, use_se=False, include_top=True): | ||
| channel_axis = 1 if K.image_data_format() == "channels_first" else -1 | ||
| if K.image_data_format() == "channels_first": | ||
| X = Input(shape=(input_shape[-1], input_shape[0], input_shape[1])) | ||
| def conv_bn_prelu(inputs, filters=-1, kernel_size=1, strides=1, padding="SAME", use_depthwise=False, use_separable=False, activation="prelu", name=""): | ||
| filters = filters if filters > 0 else inputs.shape[-1] | ||
| if use_depthwise: | ||
| nn = layers.DepthwiseConv2D(kernel_size, strides=strides, padding=padding, use_bias=False, name=name + "depthwise")(inputs) | ||
| elif use_separable: | ||
| nn = layers.SeparableConv2D(filters, kernel_size, strides=strides, padding="same", use_bias=False, name=name + "separable")(inputs) | ||
| # depthwise = layers.DepthwiseConv2D(kernel_size, strides=strides, padding=padding, use_bias=False, name=name + "depthwise")(inputs) | ||
| # nn = layers.Conv2D(filters, kernel_size=1, strides=1, padding="VALID", use_bias=False, name=name + "pointwise")(depthwise) | ||
| else: | ||
| X = Input(shape=input_shape) | ||
| M = conv_block(X, 64, 3, 2, "same") # Output Shape: (56, 56, 64) | ||
| M = separable_conv_block(M, 64, 3, 1) # (56, 56, 64) | ||
| M = inverted_residual_block(M, 64, 3, t=2, strides=2, n=5, se=use_se) # (28, 28, 64) | ||
| M = inverted_residual_block(M, 128, 3, t=4, strides=2, n=1, se=use_se) # (14, 14, 128) | ||
| M = inverted_residual_block(M, 128, 3, t=2, strides=1, n=6, se=use_se) # (14, 14, 128) | ||
| M = inverted_residual_block(M, 128, 3, t=4, strides=2, n=1, se=use_se) # (7, 7, 128) | ||
| M = inverted_residual_block(M, 128, 3, t=2, strides=1, n=2, se=use_se) # (7, 7, 128) | ||
| nn = layers.Conv2D(filters, kernel_size, strides=strides, padding=padding, use_bias=False, name=name + "conv")(inputs) | ||
|
|
||
| nn = layers.BatchNormalization(name=name + "bn")(nn) | ||
|
|
||
| if activation is not None and activation.lower() == "prelu": | ||
| nn = layers.PReLU(shared_axes=[1, 2], alpha_initializer=initializers.Constant(0.25), name=name + "prelu")(nn) | ||
| elif activation is not None: | ||
| nn = layers.Activation(activation=activation, name=name + activation)(nn) | ||
| return nn | ||
|
|
||
|
|
||
| def bottleneck(inputs, filters, expand_ratio=1, kernel_size=1, strides=1, use_residual=False, use_se=False, name=""): | ||
| hidden_channels = int(inputs.shape[-1] * expand_ratio) | ||
|
|
||
| nn = conv_bn_prelu(inputs, hidden_channels, name=name + "1_") | ||
| nn = conv_bn_prelu(nn, kernel_size=kernel_size, strides=strides, use_depthwise=True, name=name + "2_") | ||
| nn = conv_bn_prelu(nn, filters, activation=None, name=name + "3_") | ||
|
|
||
| nn = se_block(nn, name=name + "se_") if use_se else nn | ||
| nn = layers.Add()([inputs, nn]) if use_residual else nn | ||
| return nn | ||
|
|
||
|
|
||
| def MobileFaceNet( | ||
| num_blocks=[5, 1, 6, 1, 2], | ||
| out_channels=[64, 128, 128, 128, 128], | ||
| strides=[2, 2, 1, 2, 1], | ||
| expand_ratios=[2, 4, 2, 4, 2], | ||
| use_se=False, | ||
| emb_shape=256, | ||
| input_shape=(112, 112, 3), | ||
| dropout=0, | ||
| pretrained=None, | ||
| include_top=False, | ||
| name="mobile_facenet", | ||
| ): | ||
| inputs = layers.Input(shape=input_shape) # (112, 112, 3) | ||
| nn = conv_bn_prelu(inputs, filters=64, kernel_size=3, strides=2, name="stem_1_") # (56, 56, 64) | ||
| nn = conv_bn_prelu(nn, filters=64, kernel_size=3, strides=1, use_separable=True, name="stem_2_") # (56, 56, 64) | ||
|
|
||
| for id, (num_block, out_channel, stride, expand_ratio) in enumerate(zip(num_blocks, out_channels, strides, expand_ratios)): | ||
| stack_name = "stack{}_".format(id + 1) | ||
| for block_id in range(num_block): | ||
| cur_strides = stride if block_id == 0 else 1 | ||
| use_residual = False if block_id == 0 else True | ||
| block_name = stack_name + "block{}_".format(block_id + 1) | ||
| nn = bottleneck(nn, out_channel, expand_ratio, kernel_size=3, strides=cur_strides, use_residual=use_residual, use_se=use_se, name=block_name) | ||
|
|
||
| if include_top: | ||
| """pointwise_conv""" | ||
| nn = conv_bn_prelu(nn, filters=512, name="header_pointwise_") | ||
|
|
||
| """ GDC """ | ||
| M = Conv2D(512, 1, use_bias=False)(M) # (7, 7, 512) | ||
| M = BatchNormalization(axis=channel_axis)(M) | ||
| M = PReLU(shared_axes=[1, 2])(M) | ||
| M = DepthwiseConv2D(int(M.shape[1]), depth_multiplier=1, use_bias=False)(M) # (1, 1, 512) | ||
| M = BatchNormalization(axis=channel_axis)(M) | ||
| nn = layers.DepthwiseConv2D(nn.shape[1], use_bias=False, name="header_gdc_depthwise")(nn) | ||
| nn = layers.BatchNormalization(name="header_gdc_bn")(nn) | ||
|
|
||
| if dropout > 0 and dropout < 1: | ||
| M = Dropout(dropout)(M) | ||
| M = Conv2D(emb_shape, 1, use_bias=False, activation=None)(M) | ||
| M = Flatten()(M) | ||
| M = BatchNormalization(axis=channel_axis, name="embedding")(M) | ||
|
|
||
| model = Model(inputs=X, outputs=M, name=name) | ||
| if weight_file: | ||
| model.load_weights(weight_file) | ||
| nn = layers.Dropout(dropout)(nn) | ||
| nn = layers.Conv2D(emb_shape, 1, use_bias=False, name="header_gdc_post_conv")(nn) | ||
| nn = layers.Flatten()(nn) | ||
| nn = layers.BatchNormalization(name="pre_embedding")(nn) | ||
| nn = layers.Activation("linear", dtype="float32", name="embedding")(nn) | ||
|
|
||
| model = models.Model(inputs=inputs, outputs=nn, name=name) | ||
| if pretrained: | ||
| model.load_weights(pretrained) | ||
| return model |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Oops, something went wrong.