- Seunghun Paik, Sunpill Kim and Jae Hong Seo
- Accepted at BMVC 2023 (Oral)
- PyTorch
- CVXOPT
- Numpy
- etc. (e.g., matplotlib, tqdm, ...)
We modified the NbNet according to the technique introduced in the supplementary material of IronMask, including perceptual loss based on angular distance. Implementation code can be found in the folder entitled NbNet
.
Also, you can train the NbNet by referring to KD
and Train NbNet
. Prior to training NbNet, you should do knowledge distillation for constructing a face recognition model for calculating the perceptual loss.
In Security Analysis on LSH-based BTPs
, we provide an examplar notebook for doing protection & attack on demo images.