search

LSIbabnikz / FaceQAN

FaceQAN: Face Image Quality Assessment Through Adversarial Noise Exploration
26 stars 1 forks source link

How to calculate FNMR and AUC for FR models? #4

Open HOMGH opened 1 year ago

HOMGH commented 1 year ago

Hi, Thanks for sharing your source codes. Based on your experiments, you have used CurricularFace as a FR model to evaluate your method. I just checked the evaluation code of CurricularFace (https://github.com/HuangYG123/CurricularFace/blob/master/evaluate.py), but it seems only the accuracy is calculated within the code. Could you please advise how you measured other metrics like AUC and FNMR? Thanks.

LSIbabnikz commented 1 year ago

We did not include our used evaluation scripts in this repository. I suggest you look at the evaluation scripts included in the MagFace repository (https://github.com/IrvingMeng/MagFace/blob/main/eval/eval_quality/eval_quality.py), as this is also what we adapted for our paper. The code can be easily changed to work with any method or model.