attack.Bpp

class Bpp[source]

Bases: BadNet

BppAttack: Stealthy and Efficient Trojan Attacks Against Deep Neural Networks via Image Quantization and Contrastive Adversarial Learning

basic sturcture for main:

  1. config args, save_path, fix random seed

  2. set the clean train data and clean test data

  3. set the device, model, criterion, optimizer, training schedule.

  4. set the backdoor image processing, Image quantization, Dithering,

  5. training with backdoor modification simultaneously, which include Contrastive Adversarial Training

  6. save attack result

attack = Bpp()
attack.attack()

Note

@InProceedings{Wang_2022_CVPR, author = {Wang, Zhenting and Zhai, Juan and Ma, Shiqing}, title = {BppAttack: Stealthy and Efficient Trojan Attacks Against Deep Neural Networks via Image Quantization and Contrastive Adversarial Learning}, booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)}, month = {June}, year = {2022}, pages = {15074-15084}}

Parameters:

  • attack (string) – name of attack, use to match the transform and set the saving prefix of path.

  • attack_target (Int) – target class No. in all2one attack

  • attack_label_trans (str) – which type of label modification in backdoor attack

  • pratio (float) – the poison rate

  • bd_yaml_path (string) – path for yaml file provide additional default attributes

  • neg_ratio (float) – negative sample ratio

  • squeeze_num (int) – floydDitherspeed’s number of Image quantization

  • dithering (bool) – Use dithering or not

  • **kwargs (optional) – Additional attributes.