├── .gitignore ├── LICENSE ├── README.md ├── main.py ├── misc.py └── models ├── AlexNet.py ├── DenseNet.py ├── GoogleNet.py ├── LeNet.py ├── ResNet.py ├── VGG.py ├── WideResNet.py └── __init__.py /.gitignore: -------------------------------------------------------------------------------- 1 | *.idea/ 2 | .pyc 3 | *.pyc 4 | *data/ 5 | *__pycache__/ 6 | -------------------------------------------------------------------------------- /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 [2018] [IvoryCandy] 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 | # pytorch-cifar10 2 | Personal practice on CIFAR10 with PyTorch
3 | Inspired by [pytorch-cifar](https://github.com/kuangliu/pytorch-cifar) by [kuangliu](https://github.com/kuangliu). 4 | 5 | ## Introduction 6 | The CIFAR-10 dataset consists of 60000 32x32 colour images in 10 classes, with 6000 images per class. 7 | There are 50000 training images and 10000 test images. 8 | 9 | The dataset is divided into five training batches and one test batch, each with 10000 images. 10 | The test batch contains exactly 1000 randomly-selected images from each class. 11 | The training batches contain the remaining images in random order, but some training batches may contain more images from one class than another. 12 | Between them, the training batches contain exactly 5000 images from each class. 13 | 14 | ## Requirement 15 | - python3.6 16 | - numpy 17 | - pytorch 0.4.0 18 | - torchvision 0.2.0 19 | 20 | ## Usage 21 | ```bash 22 | python3 main.py 23 | ``` 24 | optional arguments: 25 | 26 | --lr default=1e-3 learning rate 27 | --epoch default=200 number of epochs tp train for 28 | --trainBatchSize default=100 training batch size 29 | --testBatchSize default=100 test batch size 30 | ## Configs 31 | __200__ epochs for each run-through,
32 | __500__ batches for each training epoch,
33 | __100__ batches for each validating epoch,
34 | __100__ images for each training and validating batch 35 | 36 | ##### Learning Rate 37 | __1e-3__ for [1,74] epochs
38 | __5e-4__ for [75,149] epochs
39 | __2.5e-4__ for [150,200) epochs
40 | 41 | ## Result 42 | Models | Accuracy | Comments 43 | :---:|:---:|:---: 44 | [LeNet](https://github.com/IvoryCandy/pytorch-cifar10/blob/master/models/LeNet.py) | 67.52% | - - - - 45 | [Alexnet](https://github.com/IvoryCandy/pytorch-cifar10/blob/master/models/AlexNet.py) | 74.74% | Result is far away from my expectation (5%+). Reasons might be inappropriate modification to fit dataset(32x32 images). 46 | [VGG11](https://github.com/IvoryCandy/pytorch-cifar10/blob/master/models/VGG.py) | 87.48% | - - - - 47 | [VGG13](https://github.com/IvoryCandy/pytorch-cifar10/blob/master/models/VGG.py) | 90.17% | - - - - 48 | [VGG16](https://github.com/IvoryCandy/pytorch-cifar10/blob/master/models/VGG.py) | TBD | - - - - 49 | [VGG19](https://github.com/IvoryCandy/pytorch-cifar10/blob/master/models/VGG.py) | TBD | - - - - 50 | [GoogleNet](https://github.com/IvoryCandy/pytorch-cifar10/blob/master/models/GoogleNet.py) | 92.57% | - - - - 51 | [ResNet18](https://github.com/IvoryCandy/pytorch-cifar10/blob/master/models/ResNet.py) | TBD | - - - - 52 | [ResNet34](https://github.com/IvoryCandy/pytorch-cifar10/blob/master/models/ResNet.py) | TBD | - - - - 53 | [ResNet50](https://github.com/IvoryCandy/pytorch-cifar10/blob/master/models/ResNet.py) | TBD | - - - - 54 | [ResNet101](https://github.com/IvoryCandy/pytorch-cifar10/blob/master/models/ResNet.py) | TBD | - - - - 55 | [ResNet152](https://github.com/IvoryCandy/pytorch-cifar10/blob/master/models/ResNet.py) | TBD | - - - - 56 | [DenseNet121](https://github.com/IvoryCandy/pytorch-cifar10/blob/master/models/DenseNet.py) | TBD | - - - - 57 | [DenseNet161](https://github.com/IvoryCandy/pytorch-cifar10/blob/master/models/DenseNet.py) | TBD | - - - - 58 | [DenseNet169](https://github.com/IvoryCandy/pytorch-cifar10/blob/master/models/DenseNet.py) | TBD | - - - - 59 | [DenseNet201](https://github.com/IvoryCandy/pytorch-cifar10/blob/master/models/DenseNet.py) | TBD | - - - - 60 | -------------------------------------------------------------------------------- /main.py: -------------------------------------------------------------------------------- 1 | import torch.optim as optim 2 | import torch.utils.data 3 | import torch.backends.cudnn as cudnn 4 | import torchvision 5 | from torchvision import transforms as transforms 6 | import numpy as np 7 | 8 | import argparse 9 | 10 | from models import * 11 | from misc import progress_bar 12 | 13 | 14 | CLASSES = ('plane', 'car', 'bird', 'cat', 'deer', 'dog', 'frog', 'horse', 'ship', 'truck') 15 | 16 | 17 | def main(): 18 | parser = argparse.ArgumentParser(description="cifar-10 with PyTorch") 19 | parser.add_argument('--lr', default=0.001, type=float, help='learning rate') 20 | parser.add_argument('--epoch', default=200, type=int, help='number of epochs tp train for') 21 | parser.add_argument('--trainBatchSize', default=100, type=int, help='training batch size') 22 | parser.add_argument('--testBatchSize', default=100, type=int, help='testing batch size') 23 | parser.add_argument('--cuda', default=torch.cuda.is_available(), type=bool, help='whether cuda is in use') 24 | args = parser.parse_args() 25 | 26 | solver = Solver(args) 27 | solver.run() 28 | 29 | 30 | class Solver(object): 31 | def __init__(self, config): 32 | self.model = None 33 | self.lr = config.lr 34 | self.epochs = config.epoch 35 | self.train_batch_size = config.trainBatchSize 36 | self.test_batch_size = config.testBatchSize 37 | self.criterion = None 38 | self.optimizer = None 39 | self.scheduler = None 40 | self.device = None 41 | self.cuda = config.cuda 42 | self.train_loader = None 43 | self.test_loader = None 44 | 45 | def load_data(self): 46 | train_transform = transforms.Compose([transforms.RandomHorizontalFlip(), transforms.ToTensor()]) 47 | test_transform = transforms.Compose([transforms.ToTensor()]) 48 | train_set = torchvision.datasets.CIFAR10(root='./data', train=True, download=True, transform=train_transform) 49 | self.train_loader = torch.utils.data.DataLoader(dataset=train_set, batch_size=self.train_batch_size, shuffle=True) 50 | test_set = torchvision.datasets.CIFAR10(root='./data', train=False, download=True, transform=test_transform) 51 | self.test_loader = torch.utils.data.DataLoader(dataset=test_set, batch_size=self.test_batch_size, shuffle=False) 52 | 53 | def load_model(self): 54 | if self.cuda: 55 | self.device = torch.device('cuda') 56 | cudnn.benchmark = True 57 | else: 58 | self.device = torch.device('cpu') 59 | 60 | # self.model = LeNet().to(self.device) 61 | # self.model = AlexNet().to(self.device) 62 | # self.model = VGG11().to(self.device) 63 | # self.model = VGG13().to(self.device) 64 | # self.model = VGG16().to(self.device) 65 | # self.model = VGG19().to(self.device) 66 | # self.model = GoogLeNet().to(self.device) 67 | # self.model = resnet18().to(self.device) 68 | # self.model = resnet34().to(self.device) 69 | # self.model = resnet50().to(self.device) 70 | # self.model = resnet101().to(self.device) 71 | # self.model = resnet152().to(self.device) 72 | # self.model = DenseNet121().to(self.device) 73 | # self.model = DenseNet161().to(self.device) 74 | # self.model = DenseNet169().to(self.device) 75 | # self.model = DenseNet201().to(self.device) 76 | self.model = WideResNet(depth=28, num_classes=10).to(self.device) 77 | 78 | self.optimizer = optim.Adam(self.model.parameters(), lr=self.lr) 79 | self.scheduler = optim.lr_scheduler.MultiStepLR(self.optimizer, milestones=[75, 150], gamma=0.5) 80 | self.criterion = nn.CrossEntropyLoss().to(self.device) 81 | 82 | def train(self): 83 | print("train:") 84 | self.model.train() 85 | train_loss = 0 86 | train_correct = 0 87 | total = 0 88 | 89 | for batch_num, (data, target) in enumerate(self.train_loader): 90 | data, target = data.to(self.device), target.to(self.device) 91 | self.optimizer.zero_grad() 92 | output = self.model(data) 93 | loss = self.criterion(output, target) 94 | loss.backward() 95 | self.optimizer.step() 96 | train_loss += loss.item() 97 | prediction = torch.max(output, 1) # second param "1" represents the dimension to be reduced 98 | total += target.size(0) 99 | 100 | # train_correct incremented by one if predicted right 101 | train_correct += np.sum(prediction[1].cpu().numpy() == target.cpu().numpy()) 102 | 103 | progress_bar(batch_num, len(self.train_loader), 'Loss: %.4f | Acc: %.3f%% (%d/%d)' 104 | % (train_loss / (batch_num + 1), 100. * train_correct / total, train_correct, total)) 105 | 106 | return train_loss, train_correct / total 107 | 108 | def test(self): 109 | print("test:") 110 | self.model.eval() 111 | test_loss = 0 112 | test_correct = 0 113 | total = 0 114 | 115 | with torch.no_grad(): 116 | for batch_num, (data, target) in enumerate(self.test_loader): 117 | data, target = data.to(self.device), target.to(self.device) 118 | output = self.model(data) 119 | loss = self.criterion(output, target) 120 | test_loss += loss.item() 121 | prediction = torch.max(output, 1) 122 | total += target.size(0) 123 | test_correct += np.sum(prediction[1].cpu().numpy() == target.cpu().numpy()) 124 | 125 | progress_bar(batch_num, len(self.test_loader), 'Loss: %.4f | Acc: %.3f%% (%d/%d)' 126 | % (test_loss / (batch_num + 1), 100. * test_correct / total, test_correct, total)) 127 | 128 | return test_loss, test_correct / total 129 | 130 | def save(self): 131 | model_out_path = "model.pth" 132 | torch.save(self.model, model_out_path) 133 | print("Checkpoint saved to {}".format(model_out_path)) 134 | 135 | def run(self): 136 | self.load_data() 137 | self.load_model() 138 | accuracy = 0 139 | for epoch in range(1, self.epochs + 1): 140 | self.scheduler.step(epoch) 141 | print("\n===> epoch: %d/200" % epoch) 142 | train_result = self.train() 143 | print(train_result) 144 | test_result = self.test() 145 | accuracy = max(accuracy, test_result[1]) 146 | if epoch == self.epochs: 147 | print("===> BEST ACC. PERFORMANCE: %.3f%%" % (accuracy * 100)) 148 | self.save() 149 | 150 | 151 | if __name__ == '__main__': 152 | main() 153 | -------------------------------------------------------------------------------- /misc.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import time 3 | 4 | 5 | TOTAL_BAR_LENGTH = 80 6 | LAST_T = time.time() 7 | BEGIN_T = LAST_T 8 | 9 | 10 | def progress_bar(current, total, msg=None): 11 | 12 | global LAST_T, BEGIN_T 13 | if current == 0: 14 | BEGIN_T = time.time() # Reset for new bar. 15 | 16 | current_len = int(TOTAL_BAR_LENGTH * (current + 1) / total) 17 | rest_len = int(TOTAL_BAR_LENGTH - current_len) - 1 18 | 19 | sys.stdout.write(' %d/%d' % (current + 1, total)) 20 | sys.stdout.write(' [') 21 | for i in range(current_len): 22 | sys.stdout.write('=') 23 | sys.stdout.write('>') 24 | for i in range(rest_len): 25 | sys.stdout.write('.') 26 | sys.stdout.write(']') 27 | 28 | current_time = time.time() 29 | step_time = current_time - LAST_T 30 | LAST_T = current_time 31 | total_time = current_time - BEGIN_T 32 | 33 | time_used = ' Step: %s' % format_time(step_time) 34 | time_used += ' | Tot: %s' % format_time(total_time) 35 | if msg: 36 | time_used += ' | ' + msg 37 | 38 | msg = time_used 39 | sys.stdout.write(msg) 40 | 41 | if current < total - 1: 42 | sys.stdout.write('\r') 43 | else: 44 | sys.stdout.write('\n') 45 | sys.stdout.flush() 46 | 47 | 48 | def format_time(seconds): 49 | days = int(seconds / 3600/24) 50 | seconds = seconds - days*3600*24 51 | hours = int(seconds / 3600) 52 | seconds = seconds - hours*3600 53 | minutes = int(seconds / 60) 54 | seconds = seconds - minutes*60 55 | secondsf = int(seconds) 56 | seconds = seconds - secondsf 57 | millis = int(seconds*1000) 58 | 59 | f = '' 60 | i = 1 61 | if days > 0: 62 | f += str(days) + 'D' 63 | i += 1 64 | if hours > 0 and i <= 2: 65 | f += str(hours) + 'h' 66 | i += 1 67 | if minutes > 0 and i <= 2: 68 | f += str(minutes) + 'm' 69 | i += 1 70 | if secondsf > 0 and i <= 2: 71 | f += str(secondsf) + 's' 72 | i += 1 73 | if millis > 0 and i <= 2: 74 | f += str(millis) + 'ms' 75 | i += 1 76 | if f == '': 77 | f = '0ms' 78 | return f 79 | -------------------------------------------------------------------------------- /models/AlexNet.py: -------------------------------------------------------------------------------- 1 | import torch.nn as nn 2 | 3 | ''' 4 | modified to fit dataset size 5 | ''' 6 | NUM_CLASSES = 10 7 | 8 | 9 | class AlexNet(nn.Module): 10 | def __init__(self, num_classes=NUM_CLASSES): 11 | super(AlexNet, self).__init__() 12 | self.features = nn.Sequential( 13 | nn.Conv2d(3, 64, kernel_size=3, stride=2, padding=1), 14 | nn.ReLU(inplace=True), 15 | nn.MaxPool2d(kernel_size=2), 16 | nn.Conv2d(64, 192, kernel_size=3, padding=1), 17 | nn.ReLU(inplace=True), 18 | nn.MaxPool2d(kernel_size=2), 19 | nn.Conv2d(192, 384, kernel_size=3, padding=1), 20 | nn.ReLU(inplace=True), 21 | nn.Conv2d(384, 256, kernel_size=3, padding=1), 22 | nn.ReLU(inplace=True), 23 | nn.Conv2d(256, 256, kernel_size=3, padding=1), 24 | nn.ReLU(inplace=True), 25 | nn.MaxPool2d(kernel_size=2), 26 | ) 27 | self.classifier = nn.Sequential( 28 | nn.Dropout(), 29 | nn.Linear(256 * 2 * 2, 4096), 30 | nn.ReLU(inplace=True), 31 | nn.Dropout(), 32 | nn.Linear(4096, 4096), 33 | nn.ReLU(inplace=True), 34 | nn.Linear(4096, num_classes), 35 | ) 36 | 37 | def forward(self, x): 38 | x = self.features(x) 39 | x = x.view(x.size(0), 256 * 2 * 2) 40 | x = self.classifier(x) 41 | return x 42 | -------------------------------------------------------------------------------- /models/DenseNet.py: -------------------------------------------------------------------------------- 1 | import math 2 | import torch 3 | import torch.nn as nn 4 | import torch.nn.functional as func 5 | 6 | 7 | class Bottleneck(nn.Module): 8 | def __init__(self, in_planes, growth_rate): 9 | super(Bottleneck, self).__init__() 10 | self.bn1 = nn.BatchNorm2d(in_planes) 11 | self.conv1 = nn.Conv2d(in_planes, 4 * growth_rate, kernel_size=1, bias=False) 12 | self.bn2 = nn.BatchNorm2d(4 * growth_rate) 13 | self.conv2 = nn.Conv2d(4 * growth_rate, growth_rate, kernel_size=3, padding=1, bias=False) 14 | 15 | def forward(self, x): 16 | y = self.conv1(func.relu(self.bn1(x))) 17 | y = self.conv2(func.relu(self.bn2(y))) 18 | x = torch.cat([y, x], 1) 19 | return x 20 | 21 | 22 | class Transition(nn.Module): 23 | def __init__(self, in_planes, out_planes): 24 | super(Transition, self).__init__() 25 | self.bn = nn.BatchNorm2d(in_planes) 26 | self.conv = nn.Conv2d(in_planes, out_planes, kernel_size=1, bias=False) 27 | 28 | def forward(self, x): 29 | x = self.conv(func.relu(self.bn(x))) 30 | x = func.avg_pool2d(x, 2) 31 | return x 32 | 33 | 34 | class DenseNet(nn.Module): 35 | def __init__(self, block, num_block, growth_rate=12, reduction=0.5, num_classes=10): 36 | super(DenseNet, self).__init__() 37 | self.growth_rate = growth_rate 38 | 39 | num_planes = 2 * growth_rate 40 | self.conv1 = nn.Conv2d(3, num_planes, kernel_size=3, padding=1, bias=False) 41 | 42 | self.dense1 = self._make_dense_layers(block, num_planes, num_block[0]) 43 | num_planes += num_block[0] * growth_rate 44 | out_planes = int(math.floor(num_planes * reduction)) 45 | self.trans1 = Transition(num_planes, out_planes) 46 | num_planes = out_planes 47 | 48 | self.dense2 = self._make_dense_layers(block, num_planes, num_block[1]) 49 | num_planes += num_block[1] * growth_rate 50 | out_planes = int(math.floor(num_planes * reduction)) 51 | self.trans2 = Transition(num_planes, out_planes) 52 | num_planes = out_planes 53 | 54 | self.dense3 = self._make_dense_layers(block, num_planes, num_block[2]) 55 | num_planes += num_block[2] * growth_rate 56 | out_planes = int(math.floor(num_planes * reduction)) 57 | self.trans3 = Transition(num_planes, out_planes) 58 | num_planes = out_planes 59 | 60 | self.dense4 = self._make_dense_layers(block, num_planes, num_block[3]) 61 | num_planes += num_block[3] * growth_rate 62 | 63 | self.bn = nn.BatchNorm2d(num_planes) 64 | self.linear = nn.Linear(num_planes, num_classes) 65 | 66 | def _make_dense_layers(self, block, in_planes, num_block): 67 | layers = [] 68 | for i in range(num_block): 69 | layers.append(block(in_planes, self.growth_rate)) 70 | in_planes += self.growth_rate 71 | return nn.Sequential(*layers) 72 | 73 | def forward(self, x): 74 | x = self.conv1(x) 75 | x = self.trans1(self.dense1(x)) 76 | x = self.trans2(self.dense2(x)) 77 | x = self.trans3(self.dense3(x)) 78 | x = self.dense4(x) 79 | x = func.avg_pool2d(func.relu(self.bn(x)), 4) 80 | x = x.view(x.size(0), -1) 81 | x = self.linear(x) 82 | return x 83 | 84 | 85 | def DenseNet121(): 86 | return DenseNet(Bottleneck, [6, 12, 24, 16], growth_rate=32) 87 | 88 | 89 | def DenseNet169(): 90 | return DenseNet(Bottleneck, [6, 12, 32, 32], growth_rate=32) 91 | 92 | 93 | def DenseNet201(): 94 | return DenseNet(Bottleneck, [6, 12, 48, 32], growth_rate=32) 95 | 96 | 97 | def DenseNet161(): 98 | return DenseNet(Bottleneck, [6, 12, 36, 24], growth_rate=48) 99 | 100 | 101 | def densenet_cifar(): 102 | return DenseNet(Bottleneck, [6, 12, 24, 16], growth_rate=12) 103 | -------------------------------------------------------------------------------- /models/GoogleNet.py: -------------------------------------------------------------------------------- 1 | import torch 2 | import torch.nn as nn 3 | 4 | 5 | class Inception(nn.Module): 6 | def __init__(self, in_planes, kernel_1_x, kernel_3_in, kernel_3_x, kernel_5_in, kernel_5_x, pool_planes): 7 | super(Inception, self).__init__() 8 | # 1x1 conv branch 9 | self.b1 = nn.Sequential( 10 | nn.Conv2d(in_planes, kernel_1_x, kernel_size=1), 11 | nn.BatchNorm2d(kernel_1_x), 12 | nn.ReLU(True), 13 | ) 14 | 15 | # 1x1 conv -> 3x3 conv branch 16 | self.b2 = nn.Sequential( 17 | nn.Conv2d(in_planes, kernel_3_in, kernel_size=1), 18 | nn.BatchNorm2d(kernel_3_in), 19 | nn.ReLU(True), 20 | nn.Conv2d(kernel_3_in, kernel_3_x, kernel_size=3, padding=1), 21 | nn.BatchNorm2d(kernel_3_x), 22 | nn.ReLU(True), 23 | ) 24 | 25 | # 1x1 conv -> 5x5 conv branch 26 | self.b3 = nn.Sequential( 27 | nn.Conv2d(in_planes, kernel_5_in, kernel_size=1), 28 | nn.BatchNorm2d(kernel_5_in), 29 | nn.ReLU(True), 30 | nn.Conv2d(kernel_5_in, kernel_5_x, kernel_size=3, padding=1), 31 | nn.BatchNorm2d(kernel_5_x), 32 | nn.ReLU(True), 33 | nn.Conv2d(kernel_5_x, kernel_5_x, kernel_size=3, padding=1), 34 | nn.BatchNorm2d(kernel_5_x), 35 | nn.ReLU(True), 36 | ) 37 | 38 | # 3x3 pool -> 1x1 conv branch 39 | self.b4 = nn.Sequential( 40 | nn.MaxPool2d(3, stride=1, padding=1), 41 | nn.Conv2d(in_planes, pool_planes, kernel_size=1), 42 | nn.BatchNorm2d(pool_planes), 43 | nn.ReLU(True), 44 | ) 45 | 46 | def forward(self, x): 47 | y1 = self.b1(x) 48 | y2 = self.b2(x) 49 | y3 = self.b3(x) 50 | y4 = self.b4(x) 51 | return torch.cat([y1,y2,y3,y4], 1) 52 | 53 | 54 | class GoogLeNet(nn.Module): 55 | def __init__(self): 56 | super(GoogLeNet, self).__init__() 57 | self.pre_layers = nn.Sequential( 58 | nn.Conv2d(3, 192, kernel_size=3, padding=1), 59 | nn.BatchNorm2d(192), 60 | nn.ReLU(True), 61 | ) 62 | 63 | self.a3 = Inception(192, 64, 96, 128, 16, 32, 32) 64 | self.b3 = Inception(256, 128, 128, 192, 32, 96, 64) 65 | 66 | self.max_pool = nn.MaxPool2d(3, stride=2, padding=1) 67 | 68 | self.a4 = Inception(480, 192, 96, 208, 16, 48, 64) 69 | self.b4 = Inception(512, 160, 112, 224, 24, 64, 64) 70 | self.c4 = Inception(512, 128, 128, 256, 24, 64, 64) 71 | self.d4 = Inception(512, 112, 144, 288, 32, 64, 64) 72 | self.e4 = Inception(528, 256, 160, 320, 32, 128, 128) 73 | 74 | self.a5 = Inception(832, 256, 160, 320, 32, 128, 128) 75 | self.b5 = Inception(832, 384, 192, 384, 48, 128, 128) 76 | 77 | self.avgpool = nn.AvgPool2d(8, stride=1) 78 | self.linear = nn.Linear(1024, 10) 79 | 80 | def forward(self, x): 81 | x = self.pre_layers(x) 82 | x = self.a3(x) 83 | x = self.b3(x) 84 | x = self.max_pool(x) 85 | x = self.a4(x) 86 | x = self.b4(x) 87 | x = self.c4(x) 88 | x = self.d4(x) 89 | x = self.e4(x) 90 | x = self.max_pool(x) 91 | x = self.a5(x) 92 | x = self.b5(x) 93 | x = self.avgpool(x) 94 | x = x.view(x.size(0), -1) 95 | x = self.linear(x) 96 | return x 97 | -------------------------------------------------------------------------------- /models/LeNet.py: -------------------------------------------------------------------------------- 1 | import torch.nn as nn 2 | import torch.nn.functional as func 3 | 4 | 5 | class LeNet(nn.Module): 6 | def __init__(self): 7 | super(LeNet, self).__init__() 8 | self.conv1 = nn.Conv2d(3, 6, kernel_size=5) 9 | self.conv2 = nn.Conv2d(6, 16, kernel_size=5) 10 | self.fc1 = nn.Linear(16*5*5, 120) 11 | self.fc2 = nn.Linear(120, 84) 12 | self.fc3 = nn.Linear(84, 10) 13 | 14 | def forward(self, x): 15 | x = func.relu(self.conv1(x)) 16 | x = func.max_pool2d(x, 2) 17 | x = func.relu(self.conv2(x)) 18 | x = func.max_pool2d(x, 2) 19 | x = x.view(x.size(0), -1) 20 | x = func.relu(self.fc1(x)) 21 | x = func.relu(self.fc2(x)) 22 | x = self.fc3(x) 23 | return x -------------------------------------------------------------------------------- /models/ResNet.py: -------------------------------------------------------------------------------- 1 | import torch.nn as nn 2 | import math 3 | 4 | 5 | def conv3x3(in_planes, out_planes, stride=1): 6 | # 3x3 convolution with padding 7 | return nn.Conv2d(in_planes, out_planes, kernel_size=3, stride=stride, padding=1, bias=False) 8 | 9 | 10 | class BasicBlock(nn.Module): 11 | expansion = 1 12 | 13 | def __init__(self, inplanes, planes, stride=1, downsample=None): 14 | super(BasicBlock, self).__init__() 15 | self.conv1 = conv3x3(inplanes, planes, stride) 16 | self.bn1 = nn.BatchNorm2d(planes) 17 | self.relu = nn.ReLU(inplace=True) 18 | self.conv2 = conv3x3(planes, planes) 19 | self.bn2 = nn.BatchNorm2d(planes) 20 | self.downsample = downsample 21 | self.stride = stride 22 | 23 | def forward(self, x): 24 | residual = x 25 | 26 | x = self.conv1(x) 27 | x = self.bn1(x) 28 | x = self.relu(x) 29 | 30 | x = self.conv2(x) 31 | x = self.bn2(x) 32 | 33 | if self.downsample is not None: 34 | residual = self.downsample(x) 35 | 36 | x += residual 37 | x = self.relu(x) 38 | 39 | return x 40 | 41 | 42 | class Bottleneck(nn.Module): 43 | expansion = 4 44 | 45 | def __init__(self, inplanes, planes, stride=1, downsample=None): 46 | super(Bottleneck, self).__init__() 47 | self.conv1 = nn.Conv2d(inplanes, planes, kernel_size=1, bias=False) 48 | self.bn1 = nn.BatchNorm2d(planes) 49 | self.conv2 = nn.Conv2d(planes, planes, kernel_size=3, stride=stride, padding=1, bias=False) 50 | self.bn2 = nn.BatchNorm2d(planes) 51 | self.conv3 = nn.Conv2d(planes, planes * 4, kernel_size=1, bias=False) 52 | self.bn3 = nn.BatchNorm2d(planes * 4) 53 | self.relu = nn.ReLU(inplace=True) 54 | self.downsample = downsample 55 | self.stride = stride 56 | 57 | def forward(self, x): 58 | residual = x 59 | 60 | x = self.conv1(x) 61 | x = self.bn1(x) 62 | x = self.relu(x) 63 | 64 | x = self.conv2(x) 65 | x = self.bn2(x) 66 | x = self.relu(x) 67 | 68 | x = self.conv3(x) 69 | x = self.bn3(x) 70 | 71 | if self.downsample is not None: 72 | residual = self.downsample(x) 73 | 74 | x += residual 75 | x = self.relu(x) 76 | 77 | return x 78 | 79 | 80 | class ResNet(nn.Module): 81 | 82 | def __init__(self, block, layers, num_classes=10): 83 | self.inplanes = 64 84 | super(ResNet, self).__init__() 85 | self.conv1 = nn.Conv2d(3, 64, kernel_size=3, stride=1, padding=1, bias=False) 86 | self.bn1 = nn.BatchNorm2d(64) 87 | self.relu = nn.ReLU(inplace=True) 88 | self.layer1 = self._make_layer(block, 64, layers[0]) 89 | self.layer2 = self._make_layer(block, 128, layers[1], stride=2) 90 | self.layer3 = self._make_layer(block, 256, layers[2], stride=2) 91 | self.layer4 = self._make_layer(block, 512, layers[3], stride=2) 92 | self.avgpool = nn.AvgPool2d(kernel_size=4) 93 | self.fc = nn.Linear(512 * block.expansion, num_classes) 94 | 95 | for m in self.modules(): 96 | if isinstance(m, nn.Conv2d): 97 | n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels 98 | m.weight.data.normal_(0, math.sqrt(2. / n)) 99 | elif isinstance(m, nn.BatchNorm2d): 100 | m.weight.data.fill_(1) 101 | m.bias.data.zero_() 102 | 103 | def _make_layer(self, block, planes, blocks, stride=1): 104 | downsample = None 105 | if stride != 1 or self.inplanes != planes * block.expansion: 106 | downsample = nn.Sequential( 107 | nn.Conv2d(self.inplanes, planes * block.expansion, kernel_size=1, stride=stride, bias=False), 108 | nn.BatchNorm2d(planes * block.expansion), 109 | ) 110 | 111 | layers = [] 112 | layers.append(block(self.inplanes, planes, stride, downsample)) 113 | self.inplanes = planes * block.expansion 114 | for i in range(1, blocks): 115 | layers.append(block(self.inplanes, planes)) 116 | return nn.Sequential(*layers) 117 | 118 | def forward(self, x): 119 | x = self.conv1(x) 120 | x = self.bn1(x) 121 | x = self.relu(x) 122 | 123 | x = self.layer1(x) 124 | x = self.layer2(x) 125 | x = self.layer3(x) 126 | x = self.layer4(x) 127 | 128 | x = self.avgpool(x) 129 | x = x.view(x.size(0), -1) 130 | x = self.fc(x) 131 | 132 | return x 133 | 134 | 135 | def resnet18(**kwargs): 136 | return ResNet(BasicBlock, [2, 2, 2, 2], **kwargs) 137 | 138 | 139 | def resnet34(**kwargs): 140 | return ResNet(BasicBlock, [3, 4, 6, 3], **kwargs) 141 | 142 | 143 | def resnet50(**kwargs): 144 | return ResNet(Bottleneck, [3, 4, 6, 3], **kwargs) 145 | 146 | 147 | def resnet101(**kwargs): 148 | return ResNet(Bottleneck, [3, 4, 23, 3], **kwargs) 149 | 150 | 151 | def resnet152(**kwargs): 152 | return ResNet(Bottleneck, [3, 8, 36, 3], **kwargs) 153 | -------------------------------------------------------------------------------- /models/VGG.py: -------------------------------------------------------------------------------- 1 | import torch.nn as nn 2 | 3 | 4 | cfg = { 5 | 'VGG11': [64, 'M', 128, 'M', 256, 256, 'M', 512, 512, 'M', 512, 512, 'M'], 6 | 'VGG13': [64, 64, 'M', 128, 128, 'M', 256, 256, 'M', 512, 512, 'M', 512, 512, 'M'], 7 | 'VGG16': [64, 64, 'M', 128, 128, 'M', 256, 256, 256, 'M', 512, 512, 512, 'M', 512, 512, 512, 'M'], 8 | 'VGG19': [64, 64, 'M', 128, 128, 'M', 256, 256, 256, 256, 'M', 512, 512, 512, 512, 'M', 512, 512, 512, 512, 'M'], 9 | } 10 | 11 | 12 | class VGG(nn.Module): 13 | def __init__(self, vgg_name): 14 | super(VGG, self).__init__() 15 | self.features = self._make_layers(cfg[vgg_name]) 16 | self.classifier = nn.Linear(512, 10) 17 | 18 | def forward(self, x): 19 | out = self.features(x) 20 | out = out.view(out.size(0), -1) 21 | out = self.classifier(out) 22 | return out 23 | 24 | def _make_layers(self, cfg): 25 | layers = [] 26 | in_channels = 3 27 | for x in cfg: 28 | if x == 'M': 29 | layers += [nn.MaxPool2d(kernel_size=2, stride=2)] 30 | else: 31 | layers += [nn.Conv2d(in_channels, x, kernel_size=3, padding=1), 32 | nn.BatchNorm2d(x), 33 | nn.ReLU(inplace=True)] 34 | in_channels = x 35 | layers += [nn.AvgPool2d(kernel_size=1, stride=1)] 36 | return nn.Sequential(*layers) 37 | 38 | 39 | def VGG11(): 40 | return VGG('VGG11') 41 | 42 | 43 | def VGG13(): 44 | return VGG('VGG13') 45 | 46 | 47 | def VGG16(): 48 | return VGG('VGG16') 49 | 50 | 51 | def VGG19(): 52 | return VGG('VGG19') 53 | -------------------------------------------------------------------------------- /models/WideResNet.py: -------------------------------------------------------------------------------- 1 | import math 2 | import torch 3 | import torch.nn as nn 4 | import torch.nn.functional as F 5 | 6 | 7 | class BasicBlock(nn.Module): 8 | def __init__(self, in_planes, out_planes, stride, drop_rate=0.0): 9 | super(BasicBlock, self).__init__() 10 | self.bn1 = nn.BatchNorm2d(in_planes) 11 | self.relu1 = nn.ReLU(inplace=True) 12 | self.conv1 = nn.Conv2d(in_planes, out_planes, kernel_size=3, stride=stride, 13 | padding=1, bias=False) 14 | self.bn2 = nn.BatchNorm2d(out_planes) 15 | self.relu2 = nn.ReLU(inplace=True) 16 | self.conv2 = nn.Conv2d(out_planes, out_planes, kernel_size=3, stride=1, 17 | padding=1, bias=False) 18 | self.droprate = drop_rate 19 | self.equalInOut = (in_planes == out_planes) 20 | self.convShortcut = (not self.equalInOut) and nn.Conv2d(in_planes, out_planes, kernel_size=1, stride=stride, 21 | padding=0, bias=False) or None 22 | 23 | def forward(self, x): 24 | if not self.equalInOut: 25 | x = self.relu1(self.bn1(x)) 26 | else: 27 | out = self.relu1(self.bn1(x)) 28 | 29 | out = self.relu2(self.bn2(self.conv1(out if self.equalInOut else x))) 30 | if self.droprate > 0: 31 | out = F.dropout(out, p=self.droprate, training=self.training) 32 | out = self.conv2(out) 33 | return torch.add(x if self.equalInOut else self.convShortcut(x), out) 34 | 35 | 36 | class NetworkBlock(nn.Module): 37 | def __init__(self, nb_layers, in_planes, out_planes, block, stride, dropRate=0.0): 38 | super(NetworkBlock, self).__init__() 39 | self.layer = self._make_layer(block, in_planes, out_planes, nb_layers, stride, dropRate) 40 | 41 | @staticmethod 42 | def _make_layer(block, in_planes, out_planes, nb_layers, stride, dropRate): 43 | layers = [] 44 | for i in range(nb_layers): 45 | layers.append(block(i == 0 and in_planes or out_planes, out_planes, i == 0 and stride or 1, dropRate)) 46 | return nn.Sequential(*layers) 47 | 48 | def forward(self, x): 49 | return self.layer(x) 50 | 51 | 52 | class WideResNet(nn.Module): 53 | def __init__(self, depth, num_classes, widen_factor=1, drop_rate=0.0): 54 | super(WideResNet, self).__init__() 55 | n_channels = [16, 16 * widen_factor, 32 * widen_factor, 64 * widen_factor] 56 | assert ((depth - 4) % 6 == 0) 57 | n = int((depth - 4) / 6) 58 | block = BasicBlock 59 | # 1st conv before any network block 60 | self.conv1 = nn.Conv2d(3, n_channels[0], kernel_size=3, stride=1, 61 | padding=1, bias=False) 62 | # 1st block 63 | self.block1 = NetworkBlock(n, n_channels[0], n_channels[1], block, 1, drop_rate) 64 | # 2nd block 65 | self.block2 = NetworkBlock(n, n_channels[1], n_channels[2], block, 2, drop_rate) 66 | # 3rd block 67 | self.block3 = NetworkBlock(n, n_channels[2], n_channels[3], block, 2, drop_rate) 68 | # global average pooling and classifier 69 | self.bn1 = nn.BatchNorm2d(n_channels[3]) 70 | self.relu = nn.ReLU(inplace=True) 71 | self.fc = nn.Linear(n_channels[3], num_classes) 72 | self.nChannels = n_channels[3] 73 | 74 | for m in self.modules(): 75 | if isinstance(m, nn.Conv2d): 76 | n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels 77 | m.weight.data.normal_(0, math.sqrt(2. / n)) 78 | elif isinstance(m, nn.BatchNorm2d): 79 | m.weight.data.fill_(1) 80 | m.bias.data.zero_() 81 | elif isinstance(m, nn.Linear): 82 | m.bias.data.zero_() 83 | 84 | def forward(self, x): 85 | out = self.conv1(x) 86 | out = self.block1(out) 87 | out = self.block2(out) 88 | out = self.block3(out) 89 | out = self.relu(self.bn1(out)) 90 | out = F.avg_pool2d(out, 8) 91 | out = out.view(-1, self.nChannels) 92 | return self.fc(out) 93 | -------------------------------------------------------------------------------- /models/__init__.py: -------------------------------------------------------------------------------- 1 | from .AlexNet import * 2 | from .VGG import * 3 | from .ResNet import * 4 | from .LeNet import * 5 | from .DenseNet import * 6 | from .GoogleNet import * 7 | from .WideResNet import * 8 | --------------------------------------------------------------------------------