Code for this paper Adversarial Feature Augmentation and Normalization for Visual Recognition. [TMLR]
Tianlong Chen, Yu Cheng, Zhe Gan, Jianfeng Wang, Lijuan Wang, Zhangyang Wang, Jingjing Liu.
Q1: Can adversarial training, as data augmentation, broadly boost the performance of various visual recognition tasks on clean data, not only image classification, but also object detection, semantic segmentation or so?
Q2: If the above answer is yes, can we have more efficient and effective options for adversarial data augmentation, e.g., avoiding the high cost of finding input-level adversarial perturbations?
More detailed results of performancing A-FAN on visual recognition tasks (e.g., classification, detection, segmentation) are referred to our paper here.
- pytorch == 1.5.0
- torchvision == 0.6.0
- advertorch
cd Classification
bash cmd/run_test.sh # for testing with a pre-trained model (AFAN model with SA 94.82%)
bash cmd/run_base.sh # for training baseline models
bash cmd/run_perturb.sh # for training ALFA models
Remark.
-
Pre-trained models can be downloaded here.
-
For classification experiments, it only support adversarial feature augmentation so far.
Detials of Faster RCNN for detection are collected here.
Detials of DeepLabv3+ for segmentation are collected here.
TBD
https://github.com/tensorflow/tpu/tree/master/models/official/efficientnet
https://github.com/lukemelas/EfficientNet-PyTorch