search

TencentYoutuResearch / Adv-Makeup

This project is an implementation of our IJCAI 2021 paper "Adv-Makeup: A New Imperceptible and Transferable Attack on Face Recognition", we propose a novel attacking form for the face recognition to synthesize the transferable and imperceptible adversarial examples.
Other
69 stars 12 forks source link

Random selection of meta-train models during training #23

Closed Werty18 closed 1 year ago

Werty18 commented 2 years ago

Hi, in paper it has been mentioned that the models will be selected randomly during meta learning (training phase). However, in model.py file, in line 264, only the first 2 models is selected every time as meta train models. Should not it be random from 3 models?

yubangji123 commented 2 years ago

This is just a training strategy in meta-learning, either you can randomly select models or select first 2 models for every iteration, which all depends on you. Note that, we keep a queue to ensure the variance of each selection in our implementation.