Fadi Boutros comments

Results 24 comments of


                                            Fadi Boutros

LFW Results not reproduced using given EUM weights and Resnet100 pre-trained weights.

Only masked embedding should be processed by EUM model. you should get exact results as in the paper

traing dataset

You should create one folder (e.g. id1, id2.....idn) for each identity in your dataset. Inside each folder, you should create two folders original (embedding of unmasked images)and fakemask (embedding of...

Underwhelimg results when testing with personal images (+ bug fix)

You need to align the testing data according to the https://github.com/fdbtrs/mixfacenets/blob/main/utils/align_trans.py

Code for inference

See https://github.com/fdbtrs/Masked-Face-Recognition-KD/issues/3#issue-1104996183

Sample code for inference

Code looks good

have erro code in verificaiton_mask.py and verification.py

Comment them out.

What is the mask tool for test datasets? from which paper or git?

https://ieeexplore.ieee.org/document/9667081

train_jonas.py is missed where to find it??

the correct file name is train.py. We update shell scriptes

Authentic training dataset used to train the model.

FFHQ

AttributeError: '_LiteModule' object has no attribute 'sample'

you need to install torch lighting < 2.0 e.g. 1.8.6. See the updated sampling instructions