• Python3.7
• paddle1.5
• cv2
• numpy
• torchvision
• torch

#!pip install -i https://pypi.tuna.tsinghua.edu.cn/simple torch

#!pip install -i https://pypi.tuna.tsinghua.edu.cn/simple torchvision

为了最大程度上模拟评测系统中的黑盒模型，训练了以下模型

• ./paddle1.5/InceptionV4
• ./paddle1.5/VGG19
• ./paddle1.5/DistResNet
• ./paddle1.5/SE_ResNeXt101_32x4d
• ./paddle1.5/SERes32x4d
• ./paddle1.6/Alex
• ./paddle1.6/DarkNet53
• ./paddle1.6/EfficientNetB4
• ./paddle1.6/Densenet161
• ./paddle1.6/DPN131
• ./paddle1.6/ResNext101_32x8d_wsl
• ./paddle1.6/SE_ResNet50_vd
• ./paddle1.6/shuffleNetV2_swish
• ./paddle1.6/Res2Net50_26w_4s
• ./paddle1.6/HRNet_W32_C（已上传）
• ./paddle1.6/ResNeXt101_vd_32x4d
• ./paddle1.6/SEnet154_vd
• ./paddle1.6/shuffleV2_x2_0
• ./paddle1.6/DenseNet264
• ./paddle1.6/HRNet_W64_C(已上传)
• ./paddle1.6/DARTS_4M
• ./paddle1.6/DARTS_6M

为了最大程度模拟灰盒模型，训练以下模型

• ./paddle1.5/ResNeXt50_32x4dxx（已上传）
• ./paddle1.5/ResNeXt50_32x4dxxx（已上传）
• ./paddle1.5/ResNeXt50_32x4dxxxx
• ./paddle1.5/ResNeXt50_32x4dxxxxx
• ./paddle1.5/ResNeXt50_32x4dxxxxxx
• ./paddle1.5/ResNeXt50_32x4dxxxxxxxxx
• ./paddle1.5/ResNeXt50_32x4dxxxxxxxxxx
• ./paddle1.5/ResNeXt50_32x4dxxxxxxxxxxx
• ./paddle1.5/ResNeXt50_32x4dxxxxxxxxxxxx
• ./paddle1.5/ResNeXt50_32x4dxxxxxxxxxxxxx

注：这些灰盒模型是经由白盒模型参数resnext50经过对抗训练得来，每次都是扩充样本后，重新训练生成的模型参数