Training the model

[Hazqeel09]

I'm trying to train the PyTorch version model with ms1mv3, I need to confirm if it will be a classification task? and how many classes that are being classified?

[Hazqeel09]

right now i have around 93K classes, is it right?

So your team train the model to classify these 93K classes using ArcLoss?

[HamadYA]

I need to confirm if it will be a classification task? and how many classes that are being classified?

Hi,

The number of classes can be obtained using:

DATA_ROOT = 'ms1m-retinaface-t1/' with open(os.path.join(DATA_ROOT, 'property'), 'r') as f: NUM_CLASS, h, w = [int(i) for i in f.read().split(',')]

Which is = 93431

The training aim, utilizing ArcFace, is to learn a similarity learning mechanism that allows distance metric learning to be solved in the classification task by introducing Angular Margin Loss.

You can check works that already implemented the training paradiam on PyTorch such as:

https://github.com/deepinsight/insightface

https://github.com/deepinsight/insightface

[Hazqeel09]

I see, thank you!

[HamadYA]

I see, thank you!

Check this too

https://github.com/fdbtrs/PocketNet

[Hazqeel09]

Will add to my list to consider, thanks