Some questions about training Inception-ResNet

[jimo17]

Hi, thanks for open source. I am very interested in your paper. I have some questions. When you use the WebFace dataset to train the Inception-ResNet network, how are the training set and test set divided?

[HanxunH]

Hi,

The dataset is randomly splited into 80% for training and 20% for testing, according to each identity.

[jimo17]

Thanks a lot for your answer. Could you please provide the Webface dataset for training Inception-ResNet in the paper?

[HanxunH]

Have a look at this thread.
happynear/AMSoftmax#18

[jimo17]

Thanks a lot for your answer. Is the image size of the Webface dataset you used when training Inception-ResNet 112*112? I look in the exp_setting.sh file, the file name of the dataset is written as casia-112x112-protected-train.

[HanxunH]

Yes, we use 112*112 for both training and testing.

[jimo17]

Thanks a lot for your answer. The Webface I downloaded is 250*250. So I would like to ask, have you processed the dataset? Or maybe I downloaded the wrong dataset?

[HanxunH]

Apologies, I lost track of where I downloaded the dataset.
Here is another pre-processed version with 112*112. I believe I downloaded the preprocessed version.
https://github.com/yule-li/CosFace/blob/master/README.md

[jimo17]

Thanks a lot for your answer. I have one more question. Does the script 'Unlearnable-Examples-main/scripts/face/min-min-noise/train_clean.sh' use clean face images to train the Inception-ResNet model?

[HanxunH]

Yes, the train_clean.sh is the baseline that is trained with the original clean dataset.

[jimo17]

Thanks a lot for your answer.​ If I want to use clean face dataset to train Inception-ResNet model according to your settings，do I need to uncomment out the following code？

https://github.com/HanxunH/Unlearnable-Examples/blob/main/main.py#L102 (contains L102-L116 )

[HanxunH]

This part is for evaluating face verifications on LFW, depending if you need the results or not.