Laser & Optoelectronics Progress, Volume. 59, Issue 16, 1615007(2022)
Deep Forgery Detection Using CutMix Algorithm and Improved Xception Network
Figures & Tables(13)
Fig. 1. CutMix augmented image example. (a)(b) Original samples; (c) augmented sample
Fig. 5. Influence of hyper-parameter α and probability p on the detection model. (a) CutMix; (b) Mixup
Table 1. XceptionNet structure
View tableTable 1. XceptionNet structure
Input size Operator Number of channels 299×299×3 Entry flow Conv1，2×2 32 149×149×32 Conv2，3×3 64 147×147×64 Block1 128 74×74×128 Block2 256 37×37×256 Block3 728 19×19×728 Middle flow Block4，3×3 728 19×19×728 Block5，3×3 728 19×19×728 Block6，3×3 728 19×19×728 Block7，3×3 728 19×19×728 Block8，3×3 728 19×19×728 Block9，3×3 728 19×19×728 Block10，3×3 728 19×19×728 Block11，3×3 728 19×19×728 Exit flow Block12 1024 10×10×1024 SeparableConv2d，3×3 1536 10×10×1536 SeparableConv2d，3×3 2048 10×10×2048 Pool，1×1
Table 2. Description of dataset
View tableTable 2. Description of dataset
Dataset Number of fake images Number of real images Train dataset 28800 7200 Test dataset 5600 1400 Validation dataset 5600 1400
Table 3. Model optimization experiment
View tableTable 3. Model optimization experiment
Description Logloss Accuracy Parameters /106 Block6 0.5539 0.8554 5.95 Block7 0.5386 0.8631 7.56 Block8 0.5258 0.8721 9.18 Block9 0.5185 0.8684 10.79 Block10 0.5398 0.8687 12.41 XceptionNet 0.5497 0.8757 20.81
Table 4. Comparison between the proposed model and other classical algorithms
View tableTable 4. Comparison between the proposed model and other classical algorithms
Method Model Logloss Accuracy Parameters /106 Method in Ref.［ 18 ］EfficientNet_b3 0.5840 0.8803 12.23 Method in Ref.［ 28 ］ResNet50 0.5413 0.8684 25.56 Method in Ref.［ 10 ］SPPNet 0.8092 0.8660 25.64 Method in Ref.［ 20 ］XceptionNet 0.5497 0.8757 20.81 Proposed method Xcep_Block8 0.5258 0.8721 9.18
Table 5. Improvements for the imbalance of sample categories
View tableTable 5. Improvements for the imbalance of sample categories
Description Logloss Accuracy Xcep_Block8 0.5258 0.8721 Xcep_Block8+Over sampling 0.4329 0.8751 XceptionNet 0.5497 0.8757 XceptionNet+Over sampling 0.4725 0.8779
Table 6. Influence of different parameter settings on hybrid data enhancement results
View tableTable 6. Influence of different parameter settings on hybrid data enhancement results
Description CutMix Mixup Logloss Accuracy Logloss Accuracy =0.5，p=0.7 0.3271 0.8760 0.3196 0.8780 =0.5，p=0.8 0.3270 0.8759 0.3080 0.8724 =0.5，p=0.9 0.3334 0.8734 0.3117 0.8737 =0.5，p=1 0.3097 0.8819 0.3170 0.8731 =1，p=0.7 0.3210 0.8743 0.3151 0.8683 =1，p=0.8 0.3376 0.8767 0.3162 0.8704 =1，p=0.9 0.3122 0.8822 0.3147 0.8686 =1，p=1 0.3185 0.8773 0.3211 0.8667
Table 7. Experimental results of data augmentation
View tableTable 7. Experimental results of data augmentation
Description Logloss Accuracy Baseline（Xcep_Block8） 0.4329 0.8751 +Cutout（size is 50） 0.5143 0.8760 +Cutout（size is 80） 0.5228 0.8750 +Cutout（size is 110） 0.4674 0.8744 +Mixup 0.3196 0.8780 +CutMix 0.3122 0.8822
Tools
Get Citation
Copy Citation Text
Pengzhi Geng, Yunqi Tang, Hongxing Fan, Xintong Zhu. Deep Forgery Detection Using CutMix Algorithm and Improved Xception Network[J]. Laser & Optoelectronics Progress, 2022, 59(16): 1615007
Paper Information
Category: Machine Vision
Received: Jul. 8, 2021
Accepted: Jul. 28, 2021
Published Online: Jul. 22, 2022
The Author Email: Tang Yunqi (tangyunqi@ppsuc.edu.cn)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20