{"id":20663666,"url":"https://github.com/vita-group/adv-ss-pretraining","last_synced_at":"2025-10-18T12:29:59.483Z","repository":{"id":107044878,"uuid":"247368554","full_name":"VITA-Group/Adv-SS-Pretraining","owner":"VITA-Group","description":"[CVPR 2020] Adversarial Robustness: From Self-Supervised Pre-Training to Fine-Tuning","archived":false,"fork":false,"pushed_at":"2021-12-30T15:26:45.000Z","size":997,"stargazers_count":85,"open_issues_count":0,"forks_count":13,"subscribers_count":13,"default_branch":"master","last_synced_at":"2025-03-29T09:41:55.970Z","etag":null,"topics":["adversarial-robustness","ensemble-pretrain","jigsaw","pre-training","rotation","self-supervised-learning","selfie"],"latest_commit_sha":null,"homepage":null,"language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":null,"status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/VITA-Group.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":null,"code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null}},"created_at":"2020-03-14T23:21:03.000Z","updated_at":"2024-12-28T08:51:09.000Z","dependencies_parsed_at":"2023-07-12T01:01:59.764Z","dependency_job_id":null,"html_url":"https://github.com/VITA-Group/Adv-SS-Pretraining","commit_stats":null,"previous_names":[],"tags_count":0,"template":false,"template_full_name":null,"repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/VITA-Group%2FAdv-SS-Pretraining","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/VITA-Group%2FAdv-SS-Pretraining/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/VITA-Group%2FAdv-SS-Pretraining/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/VITA-Group%2FAdv-SS-Pretraining/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/VITA-Group","download_url":"https://codeload.github.com/VITA-Group/Adv-SS-Pretraining/tar.gz/refs/heads/master","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":249731218,"owners_count":21317341,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2022-07-04T15:15:14.044Z","host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["adversarial-robustness","ensemble-pretrain","jigsaw","pre-training","rotation","self-supervised-learning","selfie"],"created_at":"2024-11-16T19:19:12.590Z","updated_at":"2025-10-18T12:29:54.445Z","avatar_url":"https://github.com/VITA-Group.png","language":"Python","funding_links":[],"categories":[],"sub_categories":[],"readme":"# Adversarial Robustness: From Self-Supervised Pre-Training to Fine-Tuning\n\n[![License: MIT](https://img.shields.io/badge/License-MIT-green.svg)](https://opensource.org/licenses/MIT)\n\n[Adversarial Robustness: From Self-Supervised Pre-Training to Fine-Tuning](https://arxiv.org/abs/2003.12862)\n\nTianlong Chen, Sijia Liu, Shiyu Chang, Yu Cheng, Lisa Amini, and Zhangyang Wang\n\nIn CVPR 2020.\n\n**[Trained Models](https://drive.google.com/drive/folders/18oY4mcK0qkcT5jzsRb-1A9R3Rzz5dEr8?usp=sharing) in Our Paper**. \n\n## Overview\n\nRobust pretrained models can benefit the subsequent fine-tuning in two ways: **i) boosting final model robustness; ii) saving the computation cost, if proceeding towards adversarial fine-tuning.** Here we attach the summary of our achieved performace on CIFAR-10.\n\n![](./doc_imgs/intro.png)\n\n## Methods\n\n![](./doc_imgs/method.png)\n\n## Training\n\nCurrent this code base works for Python version \u003e= 3.5, pytorch \u003e= 1.2.0, torchvision \u003e= 0.4.0\n\n**Selfie pretraining:**\n\n```shell\npython train_adv_selfie.py --gpu 0 --data -b 128 --dataset cifar --modeldir save_cifar_selfie --lr 0.1\npython train_std_selfie.py --gpu 0 --data -b 128 --dataset cifar --modeldir save_cifar_selfie --lr 0.1\n```\n\n**Rotation pretraining:**\n\n```shell\npython train_adv_rotation.py --gpu 1 --data -b 128 --save_dir adv_rotation_pretrain  --seed 22 \npython train_std_rotation.py --gpu 1 --data -b 128 --save_dir adv_rotation_pretrain  --seed 22 \n```\n\n**Jigsaw pretraining :**\n\n```shell\npython train_adv_jigsaw.py --gpu 0 --data -b 128 --save_dir adv_jigsaw_pretrain --class_number 31 --seed 22 \npython train_std_jigsaw.py --gpu 0 --data -b 128 --save_dir adv_jigsaw_pretrain --class_number 31 --seed 22 \n```\n\n**Ensemble pretrain with penalty:**\n\n```shell\npython -u ensemble_pretrain.py --gpu=1 --save_dir ensemble_pre_penalty --data ../../../ --batch_size 32\n```\n\n**Finetune:**\n\nWe offer main.py (mardy), main_trades.py and main_trades2.py three schemes for fine-tuning.\n\n```shell\npython main.py --data --batch_size --pretrained_model --save_dir --gpu\n```\n\n## Details of files\n\n### Pre-training\n\n- attack_algo.py: including the attack functions for jigsaw, rotation, selfie respectively\n- attack_algo_ensemble.py: attack function of ensemble pre-training\n- dataset.py: dataset for cifar \u0026 imagenet32\n- ensemble_pretrain.py: main code of ensemble pretrain with penalty\n- functions.py: functions for plotting\n- model_ensemble.py: model for ensemble pre-training\n- resenetv2.py: ResNet50v2\n- train_adv_jigsaw.py: main code of  adversarial jigsaw pre-training\n- train_adv_rotation.py: main code of adversarial rotation pre-training\n- train_adv_selfie.py: main code of adversarial selfie pre-training\n- train_std_jigsaw.py: main code of  standard jigsaw pre-training\n- train_std_rotation.py: main code of standard rotation pre-training\n- train_std_selfie.py: main code of standard selfie pre-training\n\n### Fine-tuning:\n\n- attack_algo.py: attack for finetune task\n- main.py: adversarial training on cifar10\n- model_ensemble.py: multi-branch model for fine-tuning\n- resnetv2.py:  Resnet50v2\n\n## Citation\n\nIf you are use this code for you research, please cite our paper.\n\n```\n@InProceedings{Chen_2020_CVPR,\nauthor = {Chen, Tianlong and Liu, Sijia and Chang, Shiyu and Cheng, Yu and Amini, Lisa and Wang, Zhangyang},\ntitle = {Adversarial Robustness: From Self-Supervised Pre-Training to Fine-Tuning},\nbooktitle = {The IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},\nmonth = {June},\nyear = {2020}\n} \n```\n\n\n\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fvita-group%2Fadv-ss-pretraining","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fvita-group%2Fadv-ss-pretraining","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fvita-group%2Fadv-ss-pretraining/lists"}