中文版 | English
In order to promote the study of data centric robust machine learning, we have opened the test set used in the preliminary / semi-finals in this competition:
Test set for stage I: https://drive.google.com/file/d/1CtK2tkYncn5uX4OJH4QQAFO_NN6ocEzy/view?usp=sharing
Test set for stage II: https://drive.google.com/file/d/1ZofA9X2cMtGC7fXA7_Qrr-1KhL3fo3u-/view?usp=sharing
The code is the training example of AAAI2022 Security AI Challenger Program Phase 8: Data Centric Robot Learning on ML models.
Contestants can quickly use the following two commands to train the wideresnet
and preactresnet18
models needed for this competition:
git clone https://github.com/vtddggg/training_template_for_AI_challenger_sea8.git && cd training_template_for_AI_challenger_sea8
sh train.sh
When you finish running, a file named Dataset.zip
will be generated in the current path. You can directly upload this file as the baseline submission.
Contestants must submit a .zip
file (including data.npy
, label.npy
, config.py
, wideresnet.pth.tar
and preactresnet18.pth.tar
). These files are generated through the following steps:
-
data.npy
,label.npy
,config.py
can be created and modified as user-defined training data and config, but they need to meet the restrictions given in here. In addition to the training data and config, other files intraining_template_for_AI_challenger_sea8
are fixed and cannot be changed by users. -
replace the above three files in
training_template_for_AI_challenger_sea8
and dosh train.sh
. -
When training is finished, submit the generated 'Dataset.zip' to the competition page.
It should be noted that after the submission evaluation stage finished, we will verify the training results of the submissions. Therefore, please always note that the wideresnet.pth.tar
and preactresnet18.pth.tar
are in the submitted Dataset.zip
are indeed generated by the corresponding data.npy
, label.npy
, config.py
.
Thank you for your participation. Wish all of you acheve good results!