├── LICENSE ├── README.md └── mobilenetv3.py /LICENSE: -------------------------------------------------------------------------------- 1 | Apache License 2 | Version 2.0, January 2004 3 | http://www.apache.org/licenses/ 4 | 5 | TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION 6 | 7 | 1. Definitions. 8 | 9 | "License" shall mean the terms and conditions for use, reproduction, 10 | and distribution as defined by Sections 1 through 9 of this document. 11 | 12 | "Licensor" shall mean the copyright owner or entity authorized by 13 | the copyright owner that is granting the License. 14 | 15 | "Legal Entity" shall mean the union of the acting entity and all 16 | other entities that control, are controlled by, or are under common 17 | control with that entity. For the purposes of this definition, 18 | "control" means (i) the power, direct or indirect, to cause the 19 | direction or management of such entity, whether by contract or 20 | otherwise, or (ii) ownership of fifty percent (50%) or more of the 21 | outstanding shares, or (iii) beneficial ownership of such entity. 22 | 23 | "You" (or "Your") shall mean an individual or Legal Entity 24 | exercising permissions granted by this License. 25 | 26 | "Source" form shall mean the preferred form for making modifications, 27 | including but not limited to software source code, documentation 28 | source, and configuration files. 29 | 30 | "Object" form shall mean any form resulting from mechanical 31 | transformation or translation of a Source form, including but 32 | not limited to compiled object code, generated documentation, 33 | and conversions to other media types. 34 | 35 | "Work" shall mean the work of authorship, whether in Source or 36 | Object form, made available under the License, as indicated by a 37 | copyright notice that is included in or attached to the work 38 | (an example is provided in the Appendix below). 39 | 40 | "Derivative Works" shall mean any work, whether in Source or Object 41 | form, that is based on (or derived from) the Work and for which the 42 | editorial revisions, annotations, elaborations, or other modifications 43 | represent, as a whole, an original work of authorship. For the purposes 44 | of this License, Derivative Works shall not include works that remain 45 | separable from, or merely link (or bind by name) to the interfaces of, 46 | the Work and Derivative Works thereof. 47 | 48 | "Contribution" shall mean any work of authorship, including 49 | the original version of the Work and any modifications or additions 50 | to that Work or Derivative Works thereof, that is intentionally 51 | submitted to Licensor for inclusion in the Work by the copyright owner 52 | or by an individual or Legal Entity authorized to submit on behalf of 53 | the copyright owner. For the purposes of this definition, "submitted" 54 | means any form of electronic, verbal, or written communication sent 55 | to the Licensor or its representatives, including but not limited to 56 | communication on electronic mailing lists, source code control systems, 57 | and issue tracking systems that are managed by, or on behalf of, the 58 | Licensor for the purpose of discussing and improving the Work, but 59 | excluding communication that is conspicuously marked or otherwise 60 | designated in writing by the copyright owner as "Not a Contribution." 61 | 62 | "Contributor" shall mean Licensor and any individual or Legal Entity 63 | on behalf of whom a Contribution has been received by Licensor and 64 | subsequently incorporated within the Work. 65 | 66 | 2. Grant of Copyright License. Subject to the terms and conditions of 67 | this License, each Contributor hereby grants to You a perpetual, 68 | worldwide, non-exclusive, no-charge, royalty-free, irrevocable 69 | copyright license to reproduce, prepare Derivative Works of, 70 | publicly display, publicly perform, sublicense, and distribute the 71 | Work and such Derivative Works in Source or Object form. 72 | 73 | 3. Grant of Patent License. Subject to the terms and conditions of 74 | this License, each Contributor hereby grants to You a perpetual, 75 | worldwide, non-exclusive, no-charge, royalty-free, irrevocable 76 | (except as stated in this section) patent license to make, have made, 77 | use, offer to sell, sell, import, and otherwise transfer the Work, 78 | where such license applies only to those patent claims licensable 79 | by such Contributor that are necessarily infringed by their 80 | Contribution(s) alone or by combination of their Contribution(s) 81 | with the Work to which such Contribution(s) was submitted. If You 82 | institute patent litigation against any entity (including a 83 | cross-claim or counterclaim in a lawsuit) alleging that the Work 84 | or a Contribution incorporated within the Work constitutes direct 85 | or contributory patent infringement, then any patent licenses 86 | granted to You under this License for that Work shall terminate 87 | as of the date such litigation is filed. 88 | 89 | 4. Redistribution. You may reproduce and distribute copies of the 90 | Work or Derivative Works thereof in any medium, with or without 91 | modifications, and in Source or Object form, provided that You 92 | meet the following conditions: 93 | 94 | (a) You must give any other recipients of the Work or 95 | Derivative Works a copy of this License; and 96 | 97 | (b) You must cause any modified files to carry prominent notices 98 | stating that You changed the files; and 99 | 100 | (c) You must retain, in the Source form of any Derivative Works 101 | that You distribute, all copyright, patent, trademark, and 102 | attribution notices from the Source form of the Work, 103 | excluding those notices that do not pertain to any part of 104 | the Derivative Works; and 105 | 106 | (d) If the Work includes a "NOTICE" text file as part of its 107 | distribution, then any Derivative Works that You distribute must 108 | include a readable copy of the attribution notices contained 109 | within such NOTICE file, excluding those notices that do not 110 | pertain to any part of the Derivative Works, in at least one 111 | of the following places: within a NOTICE text file distributed 112 | as part of the Derivative Works; within the Source form or 113 | documentation, if provided along with the Derivative Works; or, 114 | within a display generated by the Derivative Works, if and 115 | wherever such third-party notices normally appear. The contents 116 | of the NOTICE file are for informational purposes only and 117 | do not modify the License. You may add Your own attribution 118 | notices within Derivative Works that You distribute, alongside 119 | or as an addendum to the NOTICE text from the Work, provided 120 | that such additional attribution notices cannot be construed 121 | as modifying the License. 122 | 123 | You may add Your own copyright statement to Your modifications and 124 | may provide additional or different license terms and conditions 125 | for use, reproduction, or distribution of Your modifications, or 126 | for any such Derivative Works as a whole, provided Your use, 127 | reproduction, and distribution of the Work otherwise complies with 128 | the conditions stated in this License. 129 | 130 | 5. Submission of Contributions. Unless You explicitly state otherwise, 131 | any Contribution intentionally submitted for inclusion in the Work 132 | by You to the Licensor shall be under the terms and conditions of 133 | this License, without any additional terms or conditions. 134 | Notwithstanding the above, nothing herein shall supersede or modify 135 | the terms of any separate license agreement you may have executed 136 | with Licensor regarding such Contributions. 137 | 138 | 6. Trademarks. This License does not grant permission to use the trade 139 | names, trademarks, service marks, or product names of the Licensor, 140 | except as required for reasonable and customary use in describing the 141 | origin of the Work and reproducing the content of the NOTICE file. 142 | 143 | 7. Disclaimer of Warranty. Unless required by applicable law or 144 | agreed to in writing, Licensor provides the Work (and each 145 | Contributor provides its Contributions) on an "AS IS" BASIS, 146 | WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or 147 | implied, including, without limitation, any warranties or conditions 148 | of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A 149 | PARTICULAR PURPOSE. You are solely responsible for determining the 150 | appropriateness of using or redistributing the Work and assume any 151 | risks associated with Your exercise of permissions under this License. 152 | 153 | 8. Limitation of Liability. In no event and under no legal theory, 154 | whether in tort (including negligence), contract, or otherwise, 155 | unless required by applicable law (such as deliberate and grossly 156 | negligent acts) or agreed to in writing, shall any Contributor be 157 | liable to You for damages, including any direct, indirect, special, 158 | incidental, or consequential damages of any character arising as a 159 | result of this License or out of the use or inability to use the 160 | Work (including but not limited to damages for loss of goodwill, 161 | work stoppage, computer failure or malfunction, or any and all 162 | other commercial damages or losses), even if such Contributor 163 | has been advised of the possibility of such damages. 164 | 165 | 9. Accepting Warranty or Additional Liability. While redistributing 166 | the Work or Derivative Works thereof, You may choose to offer, 167 | and charge a fee for, acceptance of support, warranty, indemnity, 168 | or other liability obligations and/or rights consistent with this 169 | License. However, in accepting such obligations, You may act only 170 | on Your own behalf and on Your sole responsibility, not on behalf 171 | of any other Contributor, and only if You agree to indemnify, 172 | defend, and hold each Contributor harmless for any liability 173 | incurred by, or claims asserted against, such Contributor by reason 174 | of your accepting any such warranty or additional liability. 175 | 176 | END OF TERMS AND CONDITIONS 177 | 178 | APPENDIX: How to apply the Apache License to your work. 179 | 180 | To apply the Apache License to your work, attach the following 181 | boilerplate notice, with the fields enclosed by brackets "[]" 182 | replaced with your own identifying information. (Don't include 183 | the brackets!) The text should be enclosed in the appropriate 184 | comment syntax for the file format. We also recommend that a 185 | file or class name and description of purpose be included on the 186 | same "printed page" as the copyright notice for easier 187 | identification within third-party archives. 188 | 189 | Copyright [yyyy] [name of copyright owner] 190 | 191 | Licensed under the Apache License, Version 2.0 (the "License"); 192 | you may not use this file except in compliance with the License. 193 | You may obtain a copy of the License at 194 | 195 | http://www.apache.org/licenses/LICENSE-2.0 196 | 197 | Unless required by applicable law or agreed to in writing, software 198 | distributed under the License is distributed on an "AS IS" BASIS, 199 | WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 200 | See the License for the specific language governing permissions and 201 | limitations under the License. 202 | -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | # A PyTorch implementation of MobileNetV3 2 | 3 | This is a PyTorch implementation of MobileNetV3 architecture as described in the paper [Searching for MobileNetV3](https://arxiv.org/pdf/1905.02244.pdf). 4 | 5 | Some details may be different from the original paper, welcome to discuss and help me figure it out. 6 | 7 | - **[NEW]** The pretrained model of small version mobilenet-v3 is online, accuracy achieves the same as paper. 8 | - **[NEW]** The paper updated on 17 May, so I renew the codes for that, but there still are some bugs. 9 | - **[NEW]** I remove the se before the global avg_pool (the paper may add it in error), and now the model size is close to paper. 10 | 11 | ## Training & Accuracy 12 | ### training setting: 13 | 14 | 1. number of epochs: 150 15 | 2. learning rate schedule: cosine learning rate, initial lr=0.05 16 | 3. weight decay: 4e-5 17 | 4. remove dropout 18 | 5. batch size: 256 19 | 20 | ### MobileNetV3 large 21 | | | Madds | Parameters | Top1-acc | Pretrained Model | 22 | | ----------- | --------- | ---------- | --------- | ------------------------------------------------------------ | 23 | | Offical 1.0 | 219 M | 5.4 M | 75.2% | - | 24 | | Offical 0.75 | 155 M | 4 M | 73.3% | - | 25 | | Ours 1.0 | 224 M | 5.48 M | 72.8% | - | 26 | | Ours 0.75 | 148 M | 3.91 M | - | - | 27 | 28 | ### MobileNetV3 small 29 | | | Madds | Parameters | Top1-acc | Pretrained Model | 30 | | ----------- | --------- | ---------- | --------- | ------------------------------------------------------------ | 31 | | Offical 1.0 | 66 M | 2.9 M | 67.4% | - | 32 | | Offical 0.75 | 44 M | 2.4 M | 65.4% | - | 33 | | Ours 1.0 | 63 M | 2.94 M | 67.4% | [[google drive](https://drive.google.com/open?id=1lCsN3kWXAu8C30bQrD2JTZ7S2v4yt23C)] | 34 | | Ours 0.75 | 46 M | 2.38 M | - | - | 35 | 36 | ## Usage 37 | Pretrained models are still training ... 38 | ```python 39 | # pytorch 1.0.1 40 | # large 41 | net_large = mobilenetv3(mode='large') 42 | # small 43 | net_small = mobilenetv3(mode='small') 44 | state_dict = torch.load('mobilenetv3_small_67.4.pth.tar') 45 | net_small.load_state_dict(state_dict) 46 | ``` 47 | 48 | ## Data Pre-processing 49 | 50 | I used the following code for data pre-processing on ImageNet: 51 | 52 | ```python 53 | normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406], 54 | std=[0.229, 0.224, 0.225]) 55 | 56 | input_size = 224 57 | train_loader = torch.utils.data.DataLoader( 58 | datasets.ImageFolder( 59 | traindir, transforms.Compose([ 60 | transforms.RandomResizedCrop(input_size), 61 | transforms.RandomHorizontalFlip(), 62 | transforms.ToTensor(), 63 | normalize, 64 | ])), 65 | batch_size=batch_size, shuffle=True, 66 | num_workers=n_worker, pin_memory=True) 67 | 68 | val_loader = torch.utils.data.DataLoader( 69 | datasets.ImageFolder(valdir, transforms.Compose([ 70 | transforms.Resize(int(input_size/0.875)), 71 | transforms.CenterCrop(input_size), 72 | transforms.ToTensor(), 73 | normalize, 74 | ])), 75 | batch_size=batch_size, shuffle=False, 76 | num_workers=n_worker, pin_memory=True) 77 | ``` 78 | 79 | -------------------------------------------------------------------------------- /mobilenetv3.py: -------------------------------------------------------------------------------- 1 | import torch 2 | import torch.nn as nn 3 | import torch.nn.functional as F 4 | 5 | 6 | __all__ = ['MobileNetV3', 'mobilenetv3'] 7 | 8 | 9 | def conv_bn(inp, oup, stride, conv_layer=nn.Conv2d, norm_layer=nn.BatchNorm2d, nlin_layer=nn.ReLU): 10 | return nn.Sequential( 11 | conv_layer(inp, oup, 3, stride, 1, bias=False), 12 | norm_layer(oup), 13 | nlin_layer(inplace=True) 14 | ) 15 | 16 | 17 | def conv_1x1_bn(inp, oup, conv_layer=nn.Conv2d, norm_layer=nn.BatchNorm2d, nlin_layer=nn.ReLU): 18 | return nn.Sequential( 19 | conv_layer(inp, oup, 1, 1, 0, bias=False), 20 | norm_layer(oup), 21 | nlin_layer(inplace=True) 22 | ) 23 | 24 | 25 | class Hswish(nn.Module): 26 | def __init__(self, inplace=True): 27 | super(Hswish, self).__init__() 28 | self.inplace = inplace 29 | 30 | def forward(self, x): 31 | return x * F.relu6(x + 3., inplace=self.inplace) / 6. 32 | 33 | 34 | class Hsigmoid(nn.Module): 35 | def __init__(self, inplace=True): 36 | super(Hsigmoid, self).__init__() 37 | self.inplace = inplace 38 | 39 | def forward(self, x): 40 | return F.relu6(x + 3., inplace=self.inplace) / 6. 41 | 42 | 43 | class SEModule(nn.Module): 44 | def __init__(self, channel, reduction=4): 45 | super(SEModule, self).__init__() 46 | self.avg_pool = nn.AdaptiveAvgPool2d(1) 47 | self.fc = nn.Sequential( 48 | nn.Linear(channel, channel // reduction, bias=False), 49 | nn.ReLU(inplace=True), 50 | nn.Linear(channel // reduction, channel, bias=False), 51 | Hsigmoid() 52 | # nn.Sigmoid() 53 | ) 54 | 55 | def forward(self, x): 56 | b, c, _, _ = x.size() 57 | y = self.avg_pool(x).view(b, c) 58 | y = self.fc(y).view(b, c, 1, 1) 59 | return x * y.expand_as(x) 60 | 61 | 62 | class Identity(nn.Module): 63 | def __init__(self, channel): 64 | super(Identity, self).__init__() 65 | 66 | def forward(self, x): 67 | return x 68 | 69 | 70 | def make_divisible(x, divisible_by=8): 71 | import numpy as np 72 | return int(np.ceil(x * 1. / divisible_by) * divisible_by) 73 | 74 | 75 | class MobileBottleneck(nn.Module): 76 | def __init__(self, inp, oup, kernel, stride, exp, se=False, nl='RE'): 77 | super(MobileBottleneck, self).__init__() 78 | assert stride in [1, 2] 79 | assert kernel in [3, 5] 80 | padding = (kernel - 1) // 2 81 | self.use_res_connect = stride == 1 and inp == oup 82 | 83 | conv_layer = nn.Conv2d 84 | norm_layer = nn.BatchNorm2d 85 | if nl == 'RE': 86 | nlin_layer = nn.ReLU # or ReLU6 87 | elif nl == 'HS': 88 | nlin_layer = Hswish 89 | else: 90 | raise NotImplementedError 91 | if se: 92 | SELayer = SEModule 93 | else: 94 | SELayer = Identity 95 | 96 | self.conv = nn.Sequential( 97 | # pw 98 | conv_layer(inp, exp, 1, 1, 0, bias=False), 99 | norm_layer(exp), 100 | nlin_layer(inplace=True), 101 | # dw 102 | conv_layer(exp, exp, kernel, stride, padding, groups=exp, bias=False), 103 | norm_layer(exp), 104 | SELayer(exp), 105 | nlin_layer(inplace=True), 106 | # pw-linear 107 | conv_layer(exp, oup, 1, 1, 0, bias=False), 108 | norm_layer(oup), 109 | ) 110 | 111 | def forward(self, x): 112 | if self.use_res_connect: 113 | return x + self.conv(x) 114 | else: 115 | return self.conv(x) 116 | 117 | 118 | class MobileNetV3(nn.Module): 119 | def __init__(self, n_class=1000, input_size=224, dropout=0.8, mode='small', width_mult=1.0): 120 | super(MobileNetV3, self).__init__() 121 | input_channel = 16 122 | last_channel = 1280 123 | if mode == 'large': 124 | # refer to Table 1 in paper 125 | mobile_setting = [ 126 | # k, exp, c, se, nl, s, 127 | [3, 16, 16, False, 'RE', 1], 128 | [3, 64, 24, False, 'RE', 2], 129 | [3, 72, 24, False, 'RE', 1], 130 | [5, 72, 40, True, 'RE', 2], 131 | [5, 120, 40, True, 'RE', 1], 132 | [5, 120, 40, True, 'RE', 1], 133 | [3, 240, 80, False, 'HS', 2], 134 | [3, 200, 80, False, 'HS', 1], 135 | [3, 184, 80, False, 'HS', 1], 136 | [3, 184, 80, False, 'HS', 1], 137 | [3, 480, 112, True, 'HS', 1], 138 | [3, 672, 112, True, 'HS', 1], 139 | [5, 672, 160, True, 'HS', 2], 140 | [5, 960, 160, True, 'HS', 1], 141 | [5, 960, 160, True, 'HS', 1], 142 | ] 143 | elif mode == 'small': 144 | # refer to Table 2 in paper 145 | mobile_setting = [ 146 | # k, exp, c, se, nl, s, 147 | [3, 16, 16, True, 'RE', 2], 148 | [3, 72, 24, False, 'RE', 2], 149 | [3, 88, 24, False, 'RE', 1], 150 | [5, 96, 40, True, 'HS', 2], 151 | [5, 240, 40, True, 'HS', 1], 152 | [5, 240, 40, True, 'HS', 1], 153 | [5, 120, 48, True, 'HS', 1], 154 | [5, 144, 48, True, 'HS', 1], 155 | [5, 288, 96, True, 'HS', 2], 156 | [5, 576, 96, True, 'HS', 1], 157 | [5, 576, 96, True, 'HS', 1], 158 | ] 159 | else: 160 | raise NotImplementedError 161 | 162 | # building first layer 163 | assert input_size % 32 == 0 164 | last_channel = make_divisible(last_channel * width_mult) if width_mult > 1.0 else last_channel 165 | self.features = [conv_bn(3, input_channel, 2, nlin_layer=Hswish)] 166 | self.classifier = [] 167 | 168 | # building mobile blocks 169 | for k, exp, c, se, nl, s in mobile_setting: 170 | output_channel = make_divisible(c * width_mult) 171 | exp_channel = make_divisible(exp * width_mult) 172 | self.features.append(MobileBottleneck(input_channel, output_channel, k, s, exp_channel, se, nl)) 173 | input_channel = output_channel 174 | 175 | # building last several layers 176 | if mode == 'large': 177 | last_conv = make_divisible(960 * width_mult) 178 | self.features.append(conv_1x1_bn(input_channel, last_conv, nlin_layer=Hswish)) 179 | self.features.append(nn.AdaptiveAvgPool2d(1)) 180 | self.features.append(nn.Conv2d(last_conv, last_channel, 1, 1, 0)) 181 | self.features.append(Hswish(inplace=True)) 182 | elif mode == 'small': 183 | last_conv = make_divisible(576 * width_mult) 184 | self.features.append(conv_1x1_bn(input_channel, last_conv, nlin_layer=Hswish)) 185 | # self.features.append(SEModule(last_conv)) # refer to paper Table2, but I think this is a mistake 186 | self.features.append(nn.AdaptiveAvgPool2d(1)) 187 | self.features.append(nn.Conv2d(last_conv, last_channel, 1, 1, 0)) 188 | self.features.append(Hswish(inplace=True)) 189 | else: 190 | raise NotImplementedError 191 | 192 | # make it nn.Sequential 193 | self.features = nn.Sequential(*self.features) 194 | 195 | # building classifier 196 | self.classifier = nn.Sequential( 197 | nn.Dropout(p=dropout), # refer to paper section 6 198 | nn.Linear(last_channel, n_class), 199 | ) 200 | 201 | self._initialize_weights() 202 | 203 | def forward(self, x): 204 | x = self.features(x) 205 | x = x.mean(3).mean(2) 206 | x = self.classifier(x) 207 | return x 208 | 209 | def _initialize_weights(self): 210 | # weight initialization 211 | for m in self.modules(): 212 | if isinstance(m, nn.Conv2d): 213 | nn.init.kaiming_normal_(m.weight, mode='fan_out') 214 | if m.bias is not None: 215 | nn.init.zeros_(m.bias) 216 | elif isinstance(m, nn.BatchNorm2d): 217 | nn.init.ones_(m.weight) 218 | nn.init.zeros_(m.bias) 219 | elif isinstance(m, nn.Linear): 220 | nn.init.normal_(m.weight, 0, 0.01) 221 | if m.bias is not None: 222 | nn.init.zeros_(m.bias) 223 | 224 | 225 | def mobilenetv3(pretrained=False, **kwargs): 226 | model = MobileNetV3(**kwargs) 227 | if pretrained: 228 | state_dict = torch.load('mobilenetv3_small_67.4.pth.tar') 229 | model.load_state_dict(state_dict, strict=True) 230 | # raise NotImplementedError 231 | return model 232 | 233 | 234 | if __name__ == '__main__': 235 | net = mobilenetv3() 236 | print('mobilenetv3:\n', net) 237 | print('Total params: %.2fM' % (sum(p.numel() for p in net.parameters())/1000000.0)) 238 | input_size=(1, 3, 224, 224) 239 | # pip install --upgrade git+https://github.com/kuan-wang/pytorch-OpCounter.git 240 | from thop import profile 241 | flops, params = profile(net, input_size=input_size) 242 | # print(flops) 243 | # print(params) 244 | print('Total params: %.2fM' % (params/1000000.0)) 245 | print('Total flops: %.2fM' % (flops/1000000.0)) 246 | x = torch.randn(input_size) 247 | out = net(x) 248 | 249 | 250 | 251 | --------------------------------------------------------------------------------