├── LICENSE ├── README.md ├── Results ├── PN_ID340_Gamma_100.0 │ ├── Adv_id340_kappa10.0_Orig3_Adv5_Delta8.png │ ├── Adv_id340_kappa10.0_Orig3_Adv5_Delta8.png.npy │ ├── Delta_id340_kappa10.0_Orig3_Adv5_Delta8.png │ ├── Delta_id340_kappa10.0_Orig3_Adv5_Delta8.png.npy │ ├── Orig_original3.png │ └── Orig_original3.png.npy └── PP_ID2953_Gamma_100.0 │ ├── Adv_id2953_kappa10.0_Orig5_Adv3_Delta5.png │ ├── Adv_id2953_kappa10.0_Orig5_Adv3_Delta5.png.npy │ ├── Delta_id2953_kappa10.0_Orig5_Adv3_Delta5.png │ ├── Delta_id2953_kappa10.0_Orig5_Adv3_Delta5.png.npy │ ├── Orig_original5.png │ └── Orig_original5.png.npy ├── Utils.py ├── aen_CEM.py ├── main.py ├── models ├── AE_codec │ ├── mnist_AE_1_decoder.h5 │ └── mnist_AE_1_decoder.json └── mnist └── setup_mnist.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 | # Contrastive-Explanation-Method 2 | Codes for reproducing the contrastive explanation in “[Explanations based on the Missing: Towards Contrastive Explanations with Pertinent Negatives](https://arxiv.org/abs/1802.07623)” 3 | 4 | 5 | To find the pertinent positive (PP) of an instance, 6 | 7 | ```shell 8 | python3 main.py -i 2953 --mode PP --kappa 10 --gamma 100 9 | ``` 10 | This would find the PP of image ID 2953 in the test images from the MNIST dataset. 11 | 12 | 13 | 14 | 15 | From left to right: the original image and the pertinent positive component. This PP in Image 2953 is sufficient to be classified as 5. 16 | 17 | To find the pertinent negative (PN) of an instance, 18 | 19 | ```shell 20 | python3 main.py -i 340 --mode PN --kappa 10 --gamma 100 21 | ``` 22 | This would find the PN of image ID 340 in the test images from the MNIST dataset. 23 | 24 | 25 | 26 | 27 | From left to right: the original image, the pertinent negative component and the image composed of the original image and PN. If we add PN to Image 340, it would be classified as 5. 28 | 29 | The argument `kappa` (confidence lebel) and `gamma` (regularization coefficient of autoencoder) are tuning parameters for the optimization setup. Both PP and PN are used to explain the model prediction results. For more details, please refer to the paper. 30 | -------------------------------------------------------------------------------- /Results/PN_ID340_Gamma_100.0/Adv_id340_kappa10.0_Orig3_Adv5_Delta8.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/IBM/Contrastive-Explanation-Method/c09c28ccf2bc0b04348963d405ed702eca0f02b6/Results/PN_ID340_Gamma_100.0/Adv_id340_kappa10.0_Orig3_Adv5_Delta8.png -------------------------------------------------------------------------------- /Results/PN_ID340_Gamma_100.0/Adv_id340_kappa10.0_Orig3_Adv5_Delta8.png.npy: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/IBM/Contrastive-Explanation-Method/c09c28ccf2bc0b04348963d405ed702eca0f02b6/Results/PN_ID340_Gamma_100.0/Adv_id340_kappa10.0_Orig3_Adv5_Delta8.png.npy -------------------------------------------------------------------------------- /Results/PN_ID340_Gamma_100.0/Delta_id340_kappa10.0_Orig3_Adv5_Delta8.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/IBM/Contrastive-Explanation-Method/c09c28ccf2bc0b04348963d405ed702eca0f02b6/Results/PN_ID340_Gamma_100.0/Delta_id340_kappa10.0_Orig3_Adv5_Delta8.png -------------------------------------------------------------------------------- /Results/PN_ID340_Gamma_100.0/Delta_id340_kappa10.0_Orig3_Adv5_Delta8.png.npy: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/IBM/Contrastive-Explanation-Method/c09c28ccf2bc0b04348963d405ed702eca0f02b6/Results/PN_ID340_Gamma_100.0/Delta_id340_kappa10.0_Orig3_Adv5_Delta8.png.npy -------------------------------------------------------------------------------- /Results/PN_ID340_Gamma_100.0/Orig_original3.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/IBM/Contrastive-Explanation-Method/c09c28ccf2bc0b04348963d405ed702eca0f02b6/Results/PN_ID340_Gamma_100.0/Orig_original3.png -------------------------------------------------------------------------------- /Results/PN_ID340_Gamma_100.0/Orig_original3.png.npy: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/IBM/Contrastive-Explanation-Method/c09c28ccf2bc0b04348963d405ed702eca0f02b6/Results/PN_ID340_Gamma_100.0/Orig_original3.png.npy -------------------------------------------------------------------------------- /Results/PP_ID2953_Gamma_100.0/Adv_id2953_kappa10.0_Orig5_Adv3_Delta5.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/IBM/Contrastive-Explanation-Method/c09c28ccf2bc0b04348963d405ed702eca0f02b6/Results/PP_ID2953_Gamma_100.0/Adv_id2953_kappa10.0_Orig5_Adv3_Delta5.png -------------------------------------------------------------------------------- /Results/PP_ID2953_Gamma_100.0/Adv_id2953_kappa10.0_Orig5_Adv3_Delta5.png.npy: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/IBM/Contrastive-Explanation-Method/c09c28ccf2bc0b04348963d405ed702eca0f02b6/Results/PP_ID2953_Gamma_100.0/Adv_id2953_kappa10.0_Orig5_Adv3_Delta5.png.npy -------------------------------------------------------------------------------- /Results/PP_ID2953_Gamma_100.0/Delta_id2953_kappa10.0_Orig5_Adv3_Delta5.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/IBM/Contrastive-Explanation-Method/c09c28ccf2bc0b04348963d405ed702eca0f02b6/Results/PP_ID2953_Gamma_100.0/Delta_id2953_kappa10.0_Orig5_Adv3_Delta5.png -------------------------------------------------------------------------------- /Results/PP_ID2953_Gamma_100.0/Delta_id2953_kappa10.0_Orig5_Adv3_Delta5.png.npy: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/IBM/Contrastive-Explanation-Method/c09c28ccf2bc0b04348963d405ed702eca0f02b6/Results/PP_ID2953_Gamma_100.0/Delta_id2953_kappa10.0_Orig5_Adv3_Delta5.png.npy -------------------------------------------------------------------------------- /Results/PP_ID2953_Gamma_100.0/Orig_original5.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/IBM/Contrastive-Explanation-Method/c09c28ccf2bc0b04348963d405ed702eca0f02b6/Results/PP_ID2953_Gamma_100.0/Orig_original5.png -------------------------------------------------------------------------------- /Results/PP_ID2953_Gamma_100.0/Orig_original5.png.npy: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/IBM/Contrastive-Explanation-Method/c09c28ccf2bc0b04348963d405ed702eca0f02b6/Results/PP_ID2953_Gamma_100.0/Orig_original5.png.npy -------------------------------------------------------------------------------- /Utils.py: -------------------------------------------------------------------------------- 1 | ## Utils.py -- Some utility functions 2 | ## 3 | ## Copyright (C) 2018, IBM Corp 4 | ## Chun-Chen Tu 5 | ## PaiShun Ting 6 | ## Pin-Yu Chen 7 | ## 8 | ## Licensed under the Apache License, Version 2.0 (the "License"); 9 | ## you may not use this file except in compliance with the License. 10 | ## You may obtain a copy of the License at 11 | ## 12 | ## http://www.apache.org/licenses/LICENSE-2.0 13 | ## 14 | ## Unless required by applicable law or agreed to in writing, software 15 | ## distributed under the License is distributed on an "AS IS" BASIS, 16 | ## WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 17 | ## See the License for the specific language governing permissions and 18 | ## limitations under the License. 19 | 20 | from keras.models import Model, model_from_json, Sequential 21 | from PIL import Image 22 | 23 | import tensorflow as tf 24 | import os 25 | import numpy as np 26 | 27 | 28 | def load_AE(codec_prefix, print_summary=False): 29 | 30 | saveFilePrefix = "models/AE_codec/" + codec_prefix + "_" 31 | 32 | decoder_model_filename = saveFilePrefix + "decoder.json" 33 | decoder_weight_filename = saveFilePrefix + "decoder.h5" 34 | 35 | if not os.path.isfile(decoder_model_filename): 36 | raise Exception("The file for decoder model does not exist:{}".format(decoder_model_filename)) 37 | json_file = open(decoder_model_filename, 'r') 38 | decoder = model_from_json(json_file.read(), custom_objects={"tf": tf}) 39 | json_file.close() 40 | 41 | if not os.path.isfile(decoder_weight_filename): 42 | raise Exception("The file for decoder weights does not exist:{}".format(decoder_weight_filename)) 43 | decoder.load_weights(decoder_weight_filename) 44 | 45 | if print_summary: 46 | print("Decoder summaries") 47 | decoder.summary() 48 | 49 | return decoder 50 | 51 | def save_img(img, name = "output.png"): 52 | 53 | np.save(name, img) 54 | fig = np.around((img + 0.5)*255) 55 | fig = fig.astype(np.uint8).squeeze() 56 | pic = Image.fromarray(fig) 57 | pic.save(name) 58 | 59 | def generate_data(data, id, target_label): 60 | inputs = [] 61 | target_vec = [] 62 | 63 | inputs.append(data.test_data[id]) 64 | target_vec.append(np.eye(data.test_labels.shape[1])[target_label]) 65 | 66 | inputs = np.array(inputs) 67 | target_vec = np.array(target_vec) 68 | 69 | return inputs, target_vec 70 | 71 | def model_prediction(model, inputs): 72 | prob = model.model.predict(inputs) 73 | predicted_class = np.argmax(prob) 74 | prob_str = np.array2string(prob).replace('\n','') 75 | return prob, predicted_class, prob_str 76 | -------------------------------------------------------------------------------- /aen_CEM.py: -------------------------------------------------------------------------------- 1 | ## aen_attack.py -- attack a network optimizing elastic-net distance with an en decision rule 2 | ## when autoencoder loss is applied 3 | ## 4 | ## Copyright (C) 2018, IBM Corp 5 | ## Chun-Chen Tu 6 | ## PaiShun Ting 7 | ## Pin-Yu Chen 8 | ## 9 | ## Licensed under the Apache License, Version 2.0 (the "License"); 10 | ## you may not use this file except in compliance with the License. 11 | ## You may obtain a copy of the License at 12 | ## 13 | ## http://www.apache.org/licenses/LICENSE-2.0 14 | ## 15 | ## Unless required by applicable law or agreed to in writing, software 16 | ## distributed under the License is distributed on an "AS IS" BASIS, 17 | ## WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 18 | ## See the License for the specific language governing permissions and 19 | ## limitations under the License. 20 | 21 | import sys 22 | import tensorflow as tf 23 | import numpy as np 24 | 25 | 26 | class AEADEN: 27 | def __init__(self, sess, model, mode, AE, batch_size, kappa, init_learning_rate, 28 | binary_search_steps, max_iterations, initial_const, beta, gamma): 29 | 30 | 31 | image_size, num_channels, nun_classes = model.image_size, model.num_channels, model.num_labels 32 | shape = (batch_size, image_size, image_size, num_channels) 33 | 34 | self.sess = sess 35 | self.INIT_LEARNING_RATE = init_learning_rate 36 | self.MAX_ITERATIONS = max_iterations 37 | self.BINARY_SEARCH_STEPS = binary_search_steps 38 | self.kappa = kappa 39 | self.init_const = initial_const 40 | self.batch_size = batch_size 41 | self.AE = AE 42 | self.mode = mode 43 | self.beta = beta 44 | self.gamma = gamma 45 | 46 | # these are variables to be more efficient in sending data to tf 47 | self.orig_img = tf.Variable(np.zeros(shape), dtype=tf.float32) 48 | self.adv_img = tf.Variable(np.zeros(shape), dtype=tf.float32) 49 | self.adv_img_s = tf.Variable(np.zeros(shape), dtype=tf.float32) 50 | self.target_lab = tf.Variable(np.zeros((batch_size,nun_classes)), dtype=tf.float32) 51 | self.const = tf.Variable(np.zeros(batch_size), dtype=tf.float32) 52 | self.global_step = tf.Variable(0.0, trainable=False) 53 | 54 | # and here's what we use to assign them 55 | self.assign_orig_img = tf.placeholder(tf.float32, shape) 56 | self.assign_adv_img = tf.placeholder(tf.float32, shape) 57 | self.assign_adv_img_s = tf.placeholder(tf.float32, shape) 58 | self.assign_target_lab = tf.placeholder(tf.float32, (batch_size,nun_classes)) 59 | self.assign_const = tf.placeholder(tf.float32, [batch_size]) 60 | 61 | 62 | """Fast Iterative Soft Thresholding""" 63 | """--------------------------------""" 64 | self.zt = tf.divide(self.global_step, self.global_step+tf.cast(3, tf.float32)) 65 | 66 | cond1 = tf.cast(tf.greater(tf.subtract(self.adv_img_s, self.orig_img),self.beta), tf.float32) 67 | cond2 = tf.cast(tf.less_equal(tf.abs(tf.subtract(self.adv_img_s,self.orig_img)),self.beta), tf.float32) 68 | cond3 = tf.cast(tf.less(tf.subtract(self.adv_img_s, self.orig_img),tf.negative(self.beta)), tf.float32) 69 | upper = tf.minimum(tf.subtract(self.adv_img_s, self.beta), tf.cast(0.5, tf.float32)) 70 | lower = tf.maximum(tf.add(self.adv_img_s, self.beta), tf.cast(-0.5, tf.float32)) 71 | self.assign_adv_img = tf.multiply(cond1,upper)+tf.multiply(cond2,self.orig_img)+tf.multiply(cond3,lower) 72 | 73 | cond4=tf.cast(tf.greater(tf.subtract( self.assign_adv_img, self.orig_img),0), tf.float32) 74 | cond5=tf.cast(tf.less_equal(tf.subtract( self.assign_adv_img,self.orig_img),0), tf.float32) 75 | if self.mode == "PP": 76 | self.assign_adv_img = tf.multiply(cond5,self.assign_adv_img)+tf.multiply(cond4,self.orig_img) 77 | elif self.mode == "PN": 78 | self.assign_adv_img = tf.multiply(cond4,self.assign_adv_img)+tf.multiply(cond5,self.orig_img) 79 | 80 | self.assign_adv_img_s = self.assign_adv_img+tf.multiply(self.zt, self.assign_adv_img-self.adv_img) 81 | cond6=tf.cast(tf.greater(tf.subtract( self.assign_adv_img_s, self.orig_img),0), tf.float32) 82 | cond7=tf.cast(tf.less_equal(tf.subtract( self.assign_adv_img_s,self.orig_img),0), tf.float32) 83 | if self.mode == "PP": 84 | self.assign_adv_img_s = tf.multiply(cond7, self.assign_adv_img_s)+tf.multiply(cond6,self.orig_img) 85 | elif self.mode == "PN": 86 | self.assign_adv_img_s = tf.multiply(cond6, self.assign_adv_img_s)+tf.multiply(cond7,self.orig_img) 87 | 88 | 89 | self.adv_updater = tf.assign(self.adv_img, self.assign_adv_img) 90 | self.adv_updater_s = tf.assign(self.adv_img_s, self.assign_adv_img_s) 91 | 92 | """--------------------------------""" 93 | # prediction BEFORE-SOFTMAX of the model 94 | self.delta_img = self.orig_img-self.adv_img 95 | self.delta_img_s = self.orig_img-self.adv_img_s 96 | if self.mode == "PP": 97 | self.ImgToEnforceLabel_Score = model.predict(self.delta_img) 98 | self.ImgToEnforceLabel_Score_s = model.predict(self.delta_img_s) 99 | elif self.mode == "PN": 100 | self.ImgToEnforceLabel_Score = model.predict(self.adv_img) 101 | self.ImgToEnforceLabel_Score_s = model.predict(self.adv_img_s) 102 | 103 | # distance to the input data 104 | self.L2_dist = tf.reduce_sum(tf.square(self.delta_img),[1,2,3]) 105 | self.L2_dist_s = tf.reduce_sum(tf.square(self.delta_img_s),[1,2,3]) 106 | self.L1_dist = tf.reduce_sum(tf.abs(self.delta_img),[1,2,3]) 107 | self.L1_dist_s = tf.reduce_sum(tf.abs(self.delta_img_s),[1,2,3]) 108 | self.EN_dist = self.L2_dist + tf.multiply(self.L1_dist, self.beta) 109 | self.EN_dist_s = self.L2_dist_s + tf.multiply(self.L1_dist_s, self.beta) 110 | 111 | # compute the probability of the label class versus the maximum other 112 | self.target_lab_score = tf.reduce_sum((self.target_lab)*self.ImgToEnforceLabel_Score,1) 113 | target_lab_score_s = tf.reduce_sum((self.target_lab)*self.ImgToEnforceLabel_Score_s,1) 114 | self.max_nontarget_lab_score = tf.reduce_max((1-self.target_lab)*self.ImgToEnforceLabel_Score - (self.target_lab*10000),1) 115 | max_nontarget_lab_score_s = tf.reduce_max((1-self.target_lab)*self.ImgToEnforceLabel_Score_s - (self.target_lab*10000),1) 116 | if self.mode == "PP": 117 | Loss_Attack = tf.maximum(0.0, self.max_nontarget_lab_score - self.target_lab_score + self.kappa) 118 | Loss_Attack_s = tf.maximum(0.0, max_nontarget_lab_score_s - target_lab_score_s + self.kappa) 119 | elif self.mode == "PN": 120 | Loss_Attack = tf.maximum(0.0, -self.max_nontarget_lab_score + self.target_lab_score + self.kappa) 121 | Loss_Attack_s = tf.maximum(0.0, -max_nontarget_lab_score_s + target_lab_score_s + self.kappa) 122 | # sum up the losses 123 | self.Loss_L1Dist = tf.reduce_sum(self.L1_dist) 124 | self.Loss_L1Dist_s = tf.reduce_sum(self.L1_dist_s) 125 | self.Loss_L2Dist = tf.reduce_sum(self.L2_dist) 126 | self.Loss_L2Dist_s = tf.reduce_sum(self.L2_dist_s) 127 | self.Loss_Attack = tf.reduce_sum(self.const*Loss_Attack) 128 | self.Loss_Attack_s = tf.reduce_sum(self.const*Loss_Attack_s) 129 | if self.mode == "PP": 130 | self.Loss_AE_Dist = self.gamma*tf.square(tf.norm(self.AE(self.delta_img)-self.delta_img)) 131 | self.Loss_AE_Dist_s = self.gamma*tf.square(tf.norm(self.AE(self.delta_img)-self.delta_img_s)) 132 | elif self.mode == "PN": 133 | self.Loss_AE_Dist = self.gamma*tf.square(tf.norm(self.AE(self.adv_img)-self.adv_img)) 134 | self.Loss_AE_Dist_s = self.gamma*tf.square(tf.norm(self.AE(self.adv_img_s)-self.adv_img_s)) 135 | 136 | self.Loss_ToOptimize = self.Loss_Attack_s + self.Loss_L2Dist_s + self.Loss_AE_Dist_s 137 | self.Loss_Overall = self.Loss_Attack + self.Loss_L2Dist + self.Loss_AE_Dist + tf.multiply(self.beta, self.Loss_L1Dist) 138 | 139 | self.learning_rate = tf.train.polynomial_decay(self.INIT_LEARNING_RATE, self.global_step, self.MAX_ITERATIONS, 0, power=0.5) 140 | optimizer = tf.train.GradientDescentOptimizer(self.learning_rate) 141 | start_vars = set(x.name for x in tf.global_variables()) 142 | self.train = optimizer.minimize(self.Loss_ToOptimize, var_list=[self.adv_img_s], global_step=self.global_step) 143 | end_vars = tf.global_variables() 144 | new_vars = [x for x in end_vars if x.name not in start_vars] 145 | 146 | # these are the variables to initialize when we run 147 | self.setup = [] 148 | self.setup.append(self.orig_img.assign(self.assign_orig_img)) 149 | self.setup.append(self.target_lab.assign(self.assign_target_lab)) 150 | self.setup.append(self.const.assign(self.assign_const)) 151 | self.setup.append(self.adv_img.assign(self.assign_adv_img)) 152 | self.setup.append(self.adv_img_s.assign(self.assign_adv_img_s)) 153 | 154 | self.init = tf.variables_initializer(var_list=[self.global_step]+[self.adv_img_s]+[self.adv_img]+new_vars) 155 | 156 | def attack(self, imgs, labs): 157 | 158 | def compare(x,y): 159 | if not isinstance(x, (float, int, np.int64)): 160 | x = np.copy(x) 161 | # x[y] -= self.kappa if self.PP else -self.kappa 162 | if self.mode == "PP": 163 | x[y] -= self.kappa 164 | elif self.mode == "PN": 165 | x[y] += self.kappa 166 | x = np.argmax(x) 167 | if self.mode == "PP": 168 | return x==y 169 | else: 170 | return x!=y 171 | 172 | batch_size = self.batch_size 173 | 174 | # set the lower and upper bounds accordingly 175 | Const_LB = np.zeros(batch_size) 176 | CONST = np.ones(batch_size)*self.init_const 177 | Const_UB = np.ones(batch_size)*1e10 178 | # the best l2, score, and image attack 179 | overall_best_dist = [1e10]*batch_size 180 | overall_best_attack = [np.zeros(imgs[0].shape)]*batch_size 181 | 182 | for binary_search_steps_idx in range(self.BINARY_SEARCH_STEPS): 183 | # completely reset adam's internal state. 184 | self.sess.run(self.init) 185 | img_batch = imgs[:batch_size] 186 | label_batch = labs[:batch_size] 187 | 188 | current_step_best_dist = [1e10]*batch_size 189 | current_step_best_score = [-1]*batch_size 190 | 191 | # set the variables so that we don't have to send them over again 192 | self.sess.run(self.setup, {self.assign_orig_img: img_batch, 193 | self.assign_target_lab: label_batch, 194 | self.assign_const: CONST, 195 | self.assign_adv_img: img_batch, 196 | self.assign_adv_img_s: img_batch}) 197 | 198 | for iteration in range(self.MAX_ITERATIONS): 199 | # perform the attack 200 | self.sess.run([self.train]) 201 | self.sess.run([self.adv_updater, self.adv_updater_s]) 202 | 203 | Loss_Overall, Loss_EN, OutputScore, adv_img = self.sess.run([self.Loss_Overall, self.EN_dist, self.ImgToEnforceLabel_Score, self.adv_img]) 204 | Loss_Attack, Loss_L2Dist, Loss_L1Dist, Loss_AE_Dist = self.sess.run([self.Loss_Attack, self.Loss_L2Dist, self.Loss_L1Dist, self.Loss_AE_Dist]) 205 | target_lab_score, max_nontarget_lab_score_s = self.sess.run([self.target_lab_score, self.max_nontarget_lab_score]) 206 | 207 | if iteration%(self.MAX_ITERATIONS//10) == 0: 208 | print("iter:{} const:{}". format(iteration, CONST)) 209 | print("Loss_Overall:{:.4f}, Loss_Attack:{:.4f}". format(Loss_Overall, Loss_Attack)) 210 | print("Loss_L2Dist:{:.4f}, Loss_L1Dist:{:.4f}, AE_loss:{}". format(Loss_L2Dist, Loss_L1Dist, Loss_AE_Dist)) 211 | print("target_lab_score:{:.4f}, max_nontarget_lab_score:{:.4f}". format(target_lab_score[0], max_nontarget_lab_score_s[0])) 212 | print("") 213 | sys.stdout.flush() 214 | 215 | for batch_idx,(the_dist, the_score, the_adv_img) in enumerate(zip(Loss_EN, OutputScore, adv_img)): 216 | if the_dist < current_step_best_dist[batch_idx] and compare(the_score, np.argmax(label_batch[batch_idx])): 217 | current_step_best_dist[batch_idx] = the_dist 218 | current_step_best_score[batch_idx] = np.argmax(the_score) 219 | if the_dist < overall_best_dist[batch_idx] and compare(the_score, np.argmax(label_batch[batch_idx])): 220 | overall_best_dist[batch_idx] = the_dist 221 | overall_best_attack[batch_idx] = the_adv_img 222 | 223 | # adjust the constant as needed 224 | for batch_idx in range(batch_size): 225 | if compare(current_step_best_score[batch_idx], np.argmax(label_batch[batch_idx])) and current_step_best_score[batch_idx] != -1: 226 | # success, divide const by two 227 | Const_UB[batch_idx] = min(Const_UB[batch_idx],CONST[batch_idx]) 228 | if Const_UB[batch_idx] < 1e9: 229 | CONST[batch_idx] = (Const_LB[batch_idx] + Const_UB[batch_idx])/2 230 | else: 231 | # failure, either multiply by 10 if no solution found yet 232 | # or do binary search with the known upper bound 233 | Const_LB[batch_idx] = max(Const_LB[batch_idx],CONST[batch_idx]) 234 | if Const_UB[batch_idx] < 1e9: 235 | CONST[batch_idx] = (Const_LB[batch_idx] + Const_UB[batch_idx])/2 236 | else: 237 | CONST[batch_idx] *= 10 238 | 239 | # return the best solution found 240 | overall_best_attack = overall_best_attack[0] 241 | return overall_best_attack.reshape((1,) + overall_best_attack.shape) 242 | -------------------------------------------------------------------------------- /main.py: -------------------------------------------------------------------------------- 1 | ## main.py -- sample code to test attack procedure 2 | ## 3 | ## Copyright (C) 2018, IBM Corp 4 | ## Chun-Chen Tu 5 | ## PaiShun Ting 6 | ## Pin-Yu Chen 7 | ## 8 | ## Licensed under the Apache License, Version 2.0 (the "License"); 9 | ## you may not use this file except in compliance with the License. 10 | ## You may obtain a copy of the License at 11 | ## 12 | ## http://www.apache.org/licenses/LICENSE-2.0 13 | ## 14 | ## Unless required by applicable law or agreed to in writing, software 15 | ## distributed under the License is distributed on an "AS IS" BASIS, 16 | ## WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 17 | ## See the License for the specific language governing permissions and 18 | ## limitations under the License. 19 | 20 | import os 21 | import sys 22 | import tensorflow as tf 23 | import numpy as np 24 | import random 25 | import time 26 | from keras.layers import Lambda 27 | from setup_mnist import MNIST, MNISTModel 28 | 29 | import Utils as util 30 | from aen_CEM import AEADEN 31 | 32 | 33 | def main(args): 34 | with tf.Session() as sess: 35 | random.seed(121) 36 | np.random.seed(1211) 37 | 38 | image_id = args['img_id'] 39 | arg_max_iter = args['maxiter'] 40 | arg_b = args['binary_steps'] 41 | arg_init_const = args['init_const'] 42 | arg_mode = args['mode'] 43 | arg_kappa = args['kappa'] 44 | arg_beta = args['beta'] 45 | arg_gamma =args['gamma'] 46 | 47 | 48 | 49 | AE_model = util.load_AE("mnist_AE_1") 50 | data, model = MNIST(), MNISTModel("models/mnist", sess, False) 51 | 52 | orig_prob, orig_class, orig_prob_str = util.model_prediction(model, np.expand_dims(data.test_data[image_id], axis=0)) 53 | target_label = orig_class 54 | print("Image:{}, infer label:{}".format(image_id, target_label)) 55 | orig_img, target = util.generate_data(data, image_id, target_label) 56 | 57 | attack = AEADEN(sess, model, mode = arg_mode, AE = AE_model, batch_size=1, kappa=arg_kappa, init_learning_rate=1e-2, 58 | binary_search_steps=arg_b, max_iterations=arg_max_iter, initial_const=arg_init_const, beta=arg_beta, gamma=arg_gamma) 59 | 60 | adv_img = attack.attack(orig_img, target) 61 | 62 | adv_prob, adv_class, adv_prob_str = util.model_prediction(model, adv_img) 63 | delta_prob, delta_class, delta_prob_str = util.model_prediction(model, orig_img-adv_img) 64 | 65 | INFO = "[INFO]id:{}, kappa:{}, Orig class:{}, Adv class:{}, Delta class: {}, Orig prob:{}, Adv prob:{}, Delta prob:{}".format(image_id, arg_kappa, orig_class, adv_class, delta_class, orig_prob_str, adv_prob_str, delta_prob_str) 66 | print(INFO) 67 | 68 | suffix = "id{}_kappa{}_Orig{}_Adv{}_Delta{}".format(image_id, arg_kappa, orig_class, adv_class, delta_class) 69 | arg_save_dir = "{}_ID{}_Gamma_{}".format(arg_mode, image_id, arg_gamma) 70 | os.system("mkdir -p Results/{}".format(arg_save_dir)) 71 | util.save_img(orig_img, "Results/{}/Orig_original{}.png".format(arg_save_dir, orig_class)) 72 | util.save_img(adv_img, "Results/{}/Adv_{}.png".format(arg_save_dir, suffix)) 73 | util.save_img(np.absolute(orig_img-adv_img)-0.5, "Results/{}/Delta_{}.png".format(arg_save_dir, suffix)) 74 | 75 | sys.stdout.flush() 76 | 77 | 78 | if __name__ == "__main__": 79 | import argparse 80 | parser = argparse.ArgumentParser() 81 | parser.add_argument("-i", "--img_id", type=int) 82 | parser.add_argument("-m", "--maxiter", type=int, default=1000) 83 | parser.add_argument("-b", "--binary_steps", type=int, default=9) 84 | parser.add_argument("-c", "--init_const", type=float, default=10.0) 85 | parser.add_argument("--mode", choices=["PN", "PP"], default="PN") 86 | parser.add_argument("--kappa", type=float, default=0) 87 | parser.add_argument("--beta", type=float, default=1e-1) 88 | parser.add_argument("--gamma", type=float, default=0) 89 | 90 | args = vars(parser.parse_args()) 91 | main(args) 92 | -------------------------------------------------------------------------------- /models/AE_codec/mnist_AE_1_decoder.h5: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/IBM/Contrastive-Explanation-Method/c09c28ccf2bc0b04348963d405ed702eca0f02b6/models/AE_codec/mnist_AE_1_decoder.h5 -------------------------------------------------------------------------------- /models/AE_codec/mnist_AE_1_decoder.json: -------------------------------------------------------------------------------- 1 | {"class_name": "Sequential", "config": [{"class_name": "Sequential", "config": [{"class_name": "Conv2D", "config": {"name": "conv2d_1", "trainable": true, "batch_input_shape": [null, 28, 28, 1], "dtype": "float32", "filters": 16, "kernel_size": [3, 3], "strides": [1, 1], "padding": "same", "data_format": "channels_last", "dilation_rate": [1, 1], "activation": "linear", "use_bias": true, "kernel_initializer": {"class_name": "VarianceScaling", "config": {"scale": 1.0, "mode": "fan_avg", "distribution": "uniform", "seed": null}}, "bias_initializer": {"class_name": "Zeros", "config": {}}, "kernel_regularizer": null, "bias_regularizer": null, "activity_regularizer": null, "kernel_constraint": null, "bias_constraint": null}}, {"class_name": "Activation", "config": {"name": "activation_1", "trainable": true, "activation": "relu"}}, {"class_name": "Conv2D", "config": {"name": "conv2d_2", "trainable": true, "filters": 16, "kernel_size": [3, 3], "strides": [1, 1], "padding": "same", "data_format": "channels_last", "dilation_rate": [1, 1], "activation": "linear", "use_bias": true, "kernel_initializer": {"class_name": "VarianceScaling", "config": {"scale": 1.0, "mode": "fan_avg", "distribution": "uniform", "seed": null}}, "bias_initializer": {"class_name": "Zeros", "config": {}}, "kernel_regularizer": null, "bias_regularizer": null, "activity_regularizer": null, "kernel_constraint": null, "bias_constraint": null}}, {"class_name": "Activation", "config": {"name": "activation_2", "trainable": true, "activation": "relu"}}, {"class_name": "MaxPooling2D", "config": {"name": "max_pooling2d_1", "trainable": true, "pool_size": [2, 2], "padding": "same", "strides": [2, 2], "data_format": "channels_last"}}, {"class_name": "Conv2D", "config": {"name": "conv2d_3", "trainable": true, "filters": 1, "kernel_size": [3, 3], "strides": [1, 1], "padding": "same", "data_format": "channels_last", "dilation_rate": [1, 1], "activation": "linear", "use_bias": true, "kernel_initializer": {"class_name": "VarianceScaling", "config": {"scale": 1.0, "mode": "fan_avg", "distribution": "uniform", "seed": null}}, "bias_initializer": {"class_name": "Zeros", "config": {}}, "kernel_regularizer": null, "bias_regularizer": null, "activity_regularizer": null, "kernel_constraint": null, "bias_constraint": null}}]}, {"class_name": "Conv2D", "config": {"name": "conv2d_4", "trainable": true, "filters": 16, "kernel_size": [3, 3], "strides": [1, 1], "padding": "same", "data_format": "channels_last", "dilation_rate": [1, 1], "activation": "linear", "use_bias": true, "kernel_initializer": {"class_name": "VarianceScaling", "config": {"scale": 1.0, "mode": "fan_avg", "distribution": "uniform", "seed": null}}, "bias_initializer": {"class_name": "Zeros", "config": {}}, "kernel_regularizer": null, "bias_regularizer": null, "activity_regularizer": null, "kernel_constraint": null, "bias_constraint": null}}, {"class_name": "Activation", "config": {"name": "activation_3", "trainable": true, "activation": "relu"}}, {"class_name": "UpSampling2D", "config": {"name": "up_sampling2d_1", "trainable": true, "size": [2, 2], "data_format": "channels_last"}}, {"class_name": "Conv2D", "config": {"name": "conv2d_5", "trainable": true, "filters": 16, "kernel_size": [3, 3], "strides": [1, 1], "padding": "same", "data_format": "channels_last", "dilation_rate": [1, 1], "activation": "linear", "use_bias": true, "kernel_initializer": {"class_name": "VarianceScaling", "config": {"scale": 1.0, "mode": "fan_avg", "distribution": "uniform", "seed": null}}, "bias_initializer": {"class_name": "Zeros", "config": {}}, "kernel_regularizer": null, "bias_regularizer": null, "activity_regularizer": null, "kernel_constraint": null, "bias_constraint": null}}, {"class_name": "Activation", "config": {"name": "activation_4", "trainable": true, "activation": "relu"}}, {"class_name": "Conv2D", "config": {"name": "conv2d_6", "trainable": true, "filters": 1, "kernel_size": [3, 3], "strides": [1, 1], "padding": "same", "data_format": "channels_last", "dilation_rate": [1, 1], "activation": "linear", "use_bias": true, "kernel_initializer": {"class_name": "VarianceScaling", "config": {"scale": 1.0, "mode": "fan_avg", "distribution": "uniform", "seed": null}}, "bias_initializer": {"class_name": "Zeros", "config": {}}, "kernel_regularizer": null, "bias_regularizer": null, "activity_regularizer": null, "kernel_constraint": null, "bias_constraint": null}}], "keras_version": "2.0.5", "backend": "tensorflow"} -------------------------------------------------------------------------------- /models/mnist: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/IBM/Contrastive-Explanation-Method/c09c28ccf2bc0b04348963d405ed702eca0f02b6/models/mnist -------------------------------------------------------------------------------- /setup_mnist.py: -------------------------------------------------------------------------------- 1 | ## setup_mnist.py -- mnist data and model loading code 2 | ## 3 | ## Copyright (C) 2016, Nicholas Carlini . 4 | ## 5 | ## This program is licenced under the BSD 2-Clause licence, 6 | ## contained in the LICENCE file in this directory. 7 | 8 | import tensorflow as tf 9 | import numpy as np 10 | import os 11 | import pickle 12 | import gzip 13 | import urllib.request 14 | 15 | from keras.models import Sequential 16 | from keras.layers import Dense, Dropout, Activation, Flatten 17 | from keras.layers import Conv2D, MaxPooling2D 18 | from keras.utils import np_utils 19 | from keras.models import load_model 20 | 21 | def extract_data(filename, num_images): 22 | with gzip.open(filename) as bytestream: 23 | bytestream.read(16) 24 | buf = bytestream.read(num_images*28*28) 25 | data = np.frombuffer(buf, dtype=np.uint8).astype(np.float32) 26 | data = (data / 255) - 0.5 27 | data = data.reshape(num_images, 28, 28, 1) 28 | return data 29 | 30 | def extract_labels(filename, num_images): 31 | with gzip.open(filename) as bytestream: 32 | bytestream.read(8) 33 | buf = bytestream.read(1 * num_images) 34 | labels = np.frombuffer(buf, dtype=np.uint8) 35 | return (np.arange(10) == labels[:, None]).astype(np.float32) 36 | 37 | class MNIST: 38 | def __init__(self): 39 | if not os.path.exists("data"): 40 | os.mkdir("data") 41 | files = ["train-images-idx3-ubyte.gz", 42 | "t10k-images-idx3-ubyte.gz", 43 | "train-labels-idx1-ubyte.gz", 44 | "t10k-labels-idx1-ubyte.gz"] 45 | for name in files: 46 | 47 | urllib.request.urlretrieve('http://yann.lecun.com/exdb/mnist/' + name, "data/"+name) 48 | 49 | train_data = extract_data("data/train-images-idx3-ubyte.gz", 60000) 50 | train_labels = extract_labels("data/train-labels-idx1-ubyte.gz", 60000) 51 | self.test_data = extract_data("data/t10k-images-idx3-ubyte.gz", 10000) 52 | self.test_labels = extract_labels("data/t10k-labels-idx1-ubyte.gz", 10000) 53 | 54 | VALIDATION_SIZE = 5000 55 | 56 | self.validation_data = train_data[:VALIDATION_SIZE, :, :, :] 57 | self.validation_labels = train_labels[:VALIDATION_SIZE] 58 | self.train_data = train_data[VALIDATION_SIZE:, :, :, :] 59 | self.train_labels = train_labels[VALIDATION_SIZE:] 60 | 61 | 62 | class MNISTModel: 63 | def __init__(self, restore = None, session=None, use_log=False): 64 | self.num_channels = 1 65 | self.image_size = 28 66 | self.num_labels = 10 67 | 68 | model = Sequential() 69 | 70 | model.add(Conv2D(32, (3, 3), 71 | input_shape=(28, 28, 1))) 72 | model.add(Activation('relu')) 73 | model.add(Conv2D(32, (3, 3))) 74 | model.add(Activation('relu')) 75 | model.add(MaxPooling2D(pool_size=(2, 2))) 76 | 77 | model.add(Conv2D(64, (3, 3))) 78 | model.add(Activation('relu')) 79 | model.add(Conv2D(64, (3, 3))) 80 | model.add(Activation('relu')) 81 | model.add(MaxPooling2D(pool_size=(2, 2))) 82 | 83 | model.add(Flatten()) 84 | model.add(Dense(200)) 85 | model.add(Activation('relu')) 86 | model.add(Dense(200)) 87 | model.add(Activation('relu')) 88 | model.add(Dense(10)) 89 | # output log probability, used for black-box attack 90 | if use_log: 91 | model.add(Activation('softmax')) 92 | if restore: 93 | model.load_weights(restore) 94 | 95 | self.model = model 96 | 97 | def predict(self, data): 98 | return self.model(data) 99 | 100 | --------------------------------------------------------------------------------