| 1 | # Licensed to the Apache Software Foundation (ASF) under one | ||
| 2 | # or more contributor license agreements. See the NOTICE file | ||
| 3 | # distributed with this work for additional information | ||
| 4 | # regarding copyright ownership. The ASF licenses this file | ||
| 5 | # to you under the Apache License, Version 2.0 (the | ||
| 6 | # "License"); you may not use this file except in compliance | ||
| 7 | # with the License. You may obtain a copy of the License at | ||
| 8 | # | ||
| 9 | # http://www.apache.org/licenses/LICENSE-2.0 | ||
| 10 | # | ||
| 11 | # Unless required by applicable law or agreed to in writing, | ||
| 12 | # software distributed under the License is distributed on an | ||
| 13 | # "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY | ||
| 14 | # KIND, either express or implied. See the License for the | ||
| 15 | # specific language governing permissions and limitations | ||
| 16 | # under the License. | ||
| 17 | |||
| 18 | use strict; | ||
| 19 | use warnings; | ||
| 20 | use AI::MXNet::Function::Parameters; | ||
| 21 | package AI::MXNet::Gluon::ModelZoo::Vision::VGG; | ||
| 22 | use AI::MXNet::Gluon::Mouse; | ||
| 23 | extends 'AI::MXNet::Gluon::HybridBlock'; | ||
| 24 | use AI::MXNet::Base; | ||
| 25 | |||
| 26 | =head1 NAME | ||
| 27 | |||
| 28 | AI::MXNet::Gluon::ModelZoo::Vision::VGG - VGG model from the "Very Deep Convolutional Networks for Large-Scale Image Recognition" | ||
| 29 | =cut | ||
| 30 | |||
| 31 | =head1 DESCRIPTION | ||
| 32 | |||
| 33 | VGG model from the "Very Deep Convolutional Networks for Large-Scale Image Recognition" | ||
| 34 | <https://arxiv.org/abs/1409.1556> paper. | ||
| 35 | |||
| 36 | Parameters | ||
| 37 | ---------- | ||
| 38 | layers : array ref of Int | ||
| 39 | Numbers of layers in each feature block. | ||
| 40 | filters : array ref of Int | ||
| 41 | Numbers of filters in each feature block. List length should match the layers. | ||
| 42 | classes : Int, default 1000 | ||
| 43 | Number of classification classes. | ||
| 44 | batch_norm : Bool, default 0 | ||
| 45 | Use batch normalization. | ||
| 46 | =cut | ||
| 47 | method python_constructor_arguments() { [qw/layers filters classes batch_norm/] } | ||
| 48 | has ['layers', | ||
| 49 | 'filters'] => (is => 'ro', isa => 'ArrayRef[Int]', required => 1); | ||
| 50 | has 'classes' => (is => 'ro', isa => 'Int', default => 1000); | ||
| 51 | has 'batch_norm' => (is => 'ro', isa => 'Bool', default => 0); | ||
| 52 | |||
| 53 | sub BUILD | ||
| 54 | { | ||
| 55 | my $self = shift; | ||
| 56 | assert(@{ $self->layers } == @{ $self->filters }); | ||
| 57 | $self->name_scope(sub { | ||
| 58 | $self->features($self->_make_features()); | ||
| 59 | $self->features->add(nn->Dense(4096, activation=>'relu', | ||
| 60 | weight_initializer=>'normal', | ||
| 61 | bias_initializer=>'zeros')); | ||
| 62 | $self->features->add(nn->Dropout(rate=>0.5)); | ||
| 63 | $self->features->add(nn->Dense(4096, activation=>'relu', | ||
| 64 | weight_initializer=>'normal', | ||
| 65 | bias_initializer=>'zeros')); | ||
| 66 | $self->features->add(nn->Dropout(rate=>0.5)); | ||
| 67 | $self->output(nn->Dense($self->classes, | ||
| 68 | weight_initializer=>'normal', | ||
| 69 | bias_initializer=>'zeros')); | ||
| 70 | }); | ||
| 71 | } | ||
| 72 | |||
| 73 | method _make_features() | ||
| 74 | { | ||
| 75 | my $featurizer = nn->HybridSequential(prefix=>''); | ||
| 76 | for(enumerate($self->layers)) | ||
| 77 | { | ||
| 78 | my ($i, $num) = @$_; | ||
| 79 | for(0..$num-1) | ||
| 80 | { | ||
| 81 | $featurizer->add( | ||
| 82 | nn->Conv2D( | ||
| 83 | $self->filters->[$i], kernel_size => 3, padding => 1, | ||
| 84 | weight_initializer => mx->init->Xavier( | ||
| 85 | rnd_type => 'gaussian', | ||
| 86 | factor_type => 'out', | ||
| 87 | magnitude => 2 | ||
| 88 | ), | ||
| 89 | bias_initializer=>'zeros' | ||
| 90 | ) | ||
| 91 | ); | ||
| 92 | if($self->batch_norm) | ||
| 93 | { | ||
| 94 | $featurizer->add(nn->BatchNorm()); | ||
| 95 | } | ||
| 96 | $featurizer->add(nn->Activation('relu')); | ||
| 97 | } | ||
| 98 | $featurizer->add(nn->MaxPool2D(strides=>2)); | ||
| 99 | } | ||
| 100 | return $featurizer; | ||
| 101 | } | ||
| 102 | |||
| 103 | method hybrid_forward(GluonClass $F, GluonInput $x) | ||
| 104 | { | ||
| 105 | $x = $self->features->($x); | ||
| 106 | $x = $self->output->($x); | ||
| 107 | return $x; | ||
| 108 | } | ||
| 109 | |||
| 110 | package AI::MXNet::Gluon::ModelZoo::Vision; | ||
| 111 | |||
| 112 | # Specification | ||
| 113 | my %vgg_spec = ( | ||
| 114 | 11 => [[1, 1, 2, 2, 2], [64, 128, 256, 512, 512]], | ||
| 115 | 13 => [[2, 2, 2, 2, 2], [64, 128, 256, 512, 512]], | ||
| 116 | 16 => [[2, 2, 3, 3, 3], [64, 128, 256, 512, 512]], | ||
| 117 | 19 => [[2, 2, 4, 4, 4], [64, 128, 256, 512, 512]] | ||
| 118 | ); | ||
| 119 | |||
| 120 | =head2 get_vgg | ||
| 121 | |||
| 122 | VGG model from the "Very Deep Convolutional Networks for Large-Scale Image Recognition" | ||
| 123 | <https://arxiv.org/abs/1409.1556> paper. | ||
| 124 | |||
| 125 | Parameters | ||
| 126 | ---------- | ||
| 127 | $num_layers : Int | ||
| 128 | Number of layers for the variant of densenet. Options are 11, 13, 16, 19. | ||
| 129 | :$pretrained : Bool, default 0 | ||
| 130 | Whether to load the pretrained weights for model. | ||
| 131 | :$ctx : AI::MXNet::Context, default AI::MXNet::Context->cpu | ||
| 132 | The context in which to load the pretrained weights. | ||
| 133 | :$root : Str, default '~/.mxnet/models' | ||
| 134 | Location for keeping the model parameters. | ||
| 135 | =cut | ||
| 136 | |||
| 137 | method get_vgg( | ||
| 138 | Int $num_layers, Bool :$pretrained=0, AI::MXNet::Context :$ctx=AI::MXNet::Context->cpu(), | ||
| 139 | Str :$root='~/.mxnet/models', Int :$classes=1000, Bool :$batch_norm=0 | ||
| 140 | ) | ||
| 141 | { | ||
| 142 | my ($layers, $filters) = @{ $vgg_spec{$num_layers} }; | ||
| 143 | my $net = AI::MXNet::Gluon::ModelZoo::Vision::VGG->new($layers, $filters, $classes, $batch_norm); | ||
| 144 | if($pretrained) | ||
| 145 | { | ||
| 146 | $net->load_parameters( | ||
| 147 | AI::MXNet::Gluon::ModelZoo::ModelStore->get_model_file( | ||
| 148 | "vgg$num_layers".($batch_norm ? '_bn' : ''), | ||
| 149 | root=>$root | ||
| 150 | ), | ||
| 151 | ctx=>$ctx | ||
| 152 | ); | ||
| 153 | } | ||
| 154 | return $net; | ||
| 155 | } | ||
| 156 | |||
| 157 | =head2 vgg11 | ||
| 158 | |||
| 159 | VGG-11 model from the "Very Deep Convolutional Networks for Large-Scale Image Recognition" | ||
| 160 | <https://arxiv.org/abs/1409.1556> paper. | ||
| 161 | |||
| 162 | Parameters | ||
| 163 | ---------- | ||
| 164 | :$pretrained : Bool, default 0 | ||
| 165 | Whether to load the pretrained weights for model. | ||
| 166 | :$ctx : AI::MXNet::Context, default AI::MXNet::Context->cpu | ||
| 167 | The context in which to load the pretrained weights. | ||
| 168 | :$root : Str, default '~/.mxnet/models' | ||
| 169 | Location for keeping the model parameters. | ||
| 170 | =cut | ||
| 171 | |||
| 172 | method vgg11(%kwargs) | ||
| 173 | { | ||
| 174 | return __PACKAGE__->get_vgg(11, %kwargs); | ||
| 175 | } | ||
| 176 | |||
| 177 | =head2 vgg13 | ||
| 178 | |||
| 179 | VGG-13 model from the "Very Deep Convolutional Networks for Large-Scale Image Recognition" | ||
| 180 | <https://arxiv.org/abs/1409.1556> paper. | ||
| 181 | |||
| 182 | Parameters | ||
| 183 | ---------- | ||
| 184 | :$pretrained : Bool, default 0 | ||
| 185 | Whether to load the pretrained weights for model. | ||
| 186 | :$ctx : AI::MXNet::Context, default AI::MXNet::Context->cpu | ||
| 187 | The context in which to load the pretrained weights. | ||
| 188 | :$root : Str, default '~/.mxnet/models' | ||
| 189 | Location for keeping the model parameters. | ||
| 190 | =cut | ||
| 191 | |||
| 192 | method vgg13(%kwargs) | ||
| 193 | { | ||
| 194 | return __PACKAGE__->get_vgg(13, %kwargs); | ||
| 195 | } | ||
| 196 | |||
| 197 | =head2 vgg16 | ||
| 198 | |||
| 199 | VGG-16 model from the "Very Deep Convolutional Networks for Large-Scale Image Recognition" | ||
| 200 | <https://arxiv.org/abs/1409.1556> paper. | ||
| 201 | |||
| 202 | Parameters | ||
| 203 | ---------- | ||
| 204 | :$pretrained : Bool, default 0 | ||
| 205 | Whether to load the pretrained weights for model. | ||
| 206 | :$ctx : AI::MXNet::Context, default AI::MXNet::Context->cpu | ||
| 207 | The context in which to load the pretrained weights. | ||
| 208 | :$root : Str, default '~/.mxnet/models' | ||
| 209 | Location for keeping the model parameters. | ||
| 210 | =cut | ||
| 211 | |||
| 212 | method vgg16(%kwargs) | ||
| 213 | { | ||
| 214 | return __PACKAGE__->get_vgg(16, %kwargs); | ||
| 215 | } | ||
| 216 | |||
| 217 | =head2 vgg19 | ||
| 218 | |||
| 219 | VGG-19 model from the "Very Deep Convolutional Networks for Large-Scale Image Recognition" | ||
| 220 | <https://arxiv.org/abs/1409.1556> paper. | ||
| 221 | |||
| 222 | Parameters | ||
| 223 | ---------- | ||
| 224 | :$pretrained : Bool, default 0 | ||
| 225 | Whether to load the pretrained weights for model. | ||
| 226 | :$ctx : AI::MXNet::Context, default AI::MXNet::Context->cpu | ||
| 227 | The context in which to load the pretrained weights. | ||
| 228 | :$root : Str, default '~/.mxnet/models' | ||
| 229 | Location for keeping the model parameters. | ||
| 230 | =cut | ||
| 231 | |||
| 232 | method vgg19(%kwargs) | ||
| 233 | { | ||
| 234 | return __PACKAGE__->get_vgg(19, %kwargs); | ||
| 235 | } | ||
| 236 | |||
| 237 | =head2 vgg11_bn | ||
| 238 | |||
| 239 | VGG-11 model with batch normalization from the "Very Deep Convolutional Networks for Large-Scale Image Recognition" | ||
| 240 | <https://arxiv.org/abs/1409.1556> paper. | ||
| 241 | |||
| 242 | Parameters | ||
| 243 | ---------- | ||
| 244 | :$pretrained : Bool, default 0 | ||
| 245 | Whether to load the pretrained weights for model. | ||
| 246 | :$ctx : AI::MXNet::Context, default AI::MXNet::Context->cpu | ||
| 247 | The context in which to load the pretrained weights. | ||
| 248 | :$root : Str, default '~/.mxnet/models' | ||
| 249 | Location for keeping the model parameters. | ||
| 250 | =cut | ||
| 251 | |||
| 252 | method vgg11_bn(%kwargs) | ||
| 253 | { | ||
| 254 | $kwargs{batch_norm} = 1; | ||
| 255 | return __PACKAGE__->get_vgg(11, %kwargs); | ||
| 256 | } | ||
| 257 | |||
| 258 | =head2 vgg13_bn | ||
| 259 | |||
| 260 | VGG-13 model with batch normalization from the "Very Deep Convolutional Networks for Large-Scale Image Recognition" | ||
| 261 | <https://arxiv.org/abs/1409.1556> paper. | ||
| 262 | |||
| 263 | Parameters | ||
| 264 | ---------- | ||
| 265 | :$pretrained : Bool, default 0 | ||
| 266 | Whether to load the pretrained weights for model. | ||
| 267 | :$ctx : AI::MXNet::Context, default AI::MXNet::Context->cpu | ||
| 268 | The context in which to load the pretrained weights. | ||
| 269 | :$root : Str, default '~/.mxnet/models' | ||
| 270 | Location for keeping the model parameters. | ||
| 271 | =cut | ||
| 272 | |||
| 273 | method vgg13_bn(%kwargs) | ||
| 274 | { | ||
| 275 | $kwargs{batch_norm} = 1; | ||
| 276 | return __PACKAGE__->get_vgg(13, %kwargs); | ||
| 277 | } | ||
| 278 | |||
| 279 | =head2 vgg16_bn | ||
| 280 | |||
| 281 | VGG-16 model with batch normalization from the "Very Deep Convolutional Networks for Large-Scale Image Recognition" | ||
| 282 | <https://arxiv.org/abs/1409.1556> paper. | ||
| 283 | |||
| 284 | Parameters | ||
| 285 | ---------- | ||
| 286 | :$pretrained : Bool, default 0 | ||
| 287 | Whether to load the pretrained weights for model. | ||
| 288 | :$ctx : AI::MXNet::Context, default AI::MXNet::Context->cpu | ||
| 289 | The context in which to load the pretrained weights. | ||
| 290 | :$root : Str, default '~/.mxnet/models' | ||
| 291 | Location for keeping the model parameters. | ||
| 292 | =cut | ||
| 293 | |||
| 294 | method vgg16_bn(%kwargs) | ||
| 295 | { | ||
| 296 | $kwargs{batch_norm} = 1; | ||
| 297 | return __PACKAGE__->get_vgg(16, %kwargs); | ||
| 298 | } | ||
| 299 | |||
| 300 | =head2 vgg19_bn | ||
| 301 | |||
| 302 | VGG-19 model with batch normalization from the "Very Deep Convolutional Networks for Large-Scale Image Recognition" | ||
| 303 | <https://arxiv.org/abs/1409.1556> paper. | ||
| 304 | |||
| 305 | Parameters | ||
| 306 | ---------- | ||
| 307 | :$pretrained : Bool, default 0 | ||
| 308 | Whether to load the pretrained weights for model. | ||
| 309 | :$ctx : AI::MXNet::Context, default AI::MXNet::Context->cpu | ||
| 310 | The context in which to load the pretrained weights. | ||
| 311 | :$root : Str, default '~/.mxnet/models' | ||
| 312 | Location for keeping the model parameters. | ||
| 313 | =cut | ||
| 314 | |||
| 315 | method vgg19_bn(%kwargs) | ||
| 316 | { | ||
| 317 | $kwargs{batch_norm} = 1; | ||
| 318 | return __PACKAGE__->get_vgg(19, %kwargs); | ||
| 319 | } | ||
| 320 | |||
| 321 | 1; |