Labeling with clustering models (Gansbeke W etal.), generating noises with error-minimizing (Huang etal.) method.
The repository is based on the Code from ICLR2021 Spotlight Paper "Unlearnable Examples: Making Personal Data Unexploitable " by Hanxun Huang, Xingjun Ma, Sarah Monazam Erfani, James Bailey, Yisen Wang.
的
- Add experiments for label-agnostic dataset .
- Supply AgnosticCIFAR10Folder etc. classes and
- Some robust Loss functions for more robust training.
- SCELoss "Symmetric Cross Entropy for Robust Learning with Noisy Labels" by Yisen Wang, Xingjun Ma, Zaiyi Chen, Yuan Luo, Jin-Feng Yi, James Bailey.
Dowmload the agnostic-label CIFAR10 dataset from here. The labels are generated by deep clustering models.
Set the argument --seed , it will automatically genarates an experiment folder.
The args --train_data_path, --test_data_path should be set to your own path.
--seed 4
--version resnet18
--exp_name result/agnostic_cifar10/min-min/samplewise/
--config_path configs/cifar10
--train_batch_size 512
--eval_batch_size 512
--num_of_workers 0
--train_data_type AgnosticCIFAR10Folder
--train_data_path C:\Users\zhangyisheng\Desktop\My-Unsupervised-Classification-master\datasets\agnostic-label-cifar-10-clean
--test_data_type CIFAR10
--test_data_path C:\Users\zhangyisheng\Desktop\My-Unsupervised-Classification-master\datasets\cifar-10
--universal_stop_error 0.01
--train_step 10
--attack_type min-min
--perturb_type samplewise
--noise_shape 50000 3 32 32
--epsilon 16
--num_steps 20
--step_size 0.8
Remenber set the arg --seed and the arg --perturb_tensor_filepath simultaneously.
--seed 4
--version resnet18
--exp_name result/agnostic_cifar10/min-min/samplewise/
--config_path configs/cifar10
--train_data_type PoisonAgnosticCIFAR10Folder
--poison_rate 1.0
--perturb_type samplewise
--perturb_tensor_filepath result/agnostic_cifar10/min-min/samplewise\resnet18_seed4/perturbation.pt
--train
--num_of_workers 0
--train_data_path C:\Users\zhangyisheng\Desktop\My-Unsupervised-Classification-master\datasets\agnostic-label-cifar-10-clean
--test_data_path C:\Users\zhangyisheng\Desktop\My-Unsupervised-Classification-master\datasets\cifar-10
--seed 4
--config_path configs/cifar10
--exp_name result/agnostic_cifar10/min-min/classwise
--version resnet18
--train_data_type AgnosticCIFAR10Folder
--noise_shape 10 3 32 32
--epsilon 16
--num_steps 1
--step_size 0.8
--attack_type min-min
--perturb_type classwise
--universal_train_target train_subset
--universal_stop_error 0.1 --use_subset
--num_of_workers 0
--train_data_path C:\Users\zhangyisheng\Desktop\My-Unsupervised-Classification-master\datasets\agnostic-label-cifar-10-clean
--test_data_path C:\Users\zhangyisheng\Desktop\My-Unsupervised-Classification-master\datasets\cifar-10
--seed 4
--version resnet18
--exp_name result/agnostic_cifar10/min-min/classwise/
--config_path configs/cifar10
--train_data_type PoisonAgnosticCIFAR10Folder
--poison_rate 1.0
--perturb_type classwise
--perturb_tensor_filepath result/agnostic_cifar10/min-min/classwise/resnet18_seed4\perturbation.pt
--train
--num_of_workers 0
--train_data_path C:\Users\zhangyisheng\Desktop\My-Unsupervised-Classification-master\datasets\agnostic-label-cifar-10-clean
--test_data_path C:\Users\zhangyisheng\Desktop\My-Unsupervised-Classification-master\datasets\cifar-10
ICLR2021 Spotlight Paper "Unlearnable Examples: Making Personal Data Unexploitable " by Hanxun Huang, Xingjun Ma, Sarah Monazam Erfani, James Bailey, Yisen Wang.
ECCV2020 "SCAN: Learning to Classify Images without Labels" by Wouter Van Gansbeke, Simon Vandenhende, Stamatios Georgoulis, Marc Proesmans, Luc van Gool.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent