reds-lab / Narcissus

The official implementation of the CCS'23 paper, Narcissus clean-label backdoor attack -- only takes THREE images to poison a face recognition dataset in a clean-label way and achieves a 99.89% attack success rate.
https://arxiv.org/pdf/2204.05255.pdf
MIT License
105 stars 12 forks source link

Segmentation fault (core dumped) #9

Open 1Opera opened 8 months ago

1Opera commented 8 months ago

在服务器运行代码时报错是因为内存不足嘛 Segmentation fault (core dumped)