Porting model weight

[asd841018]

Hi, thank you for the awesome work.

I notice that in your Ported Models (e.g. [r100 12m AdaFace]) description, you ported the original repo's weight to tensorflow version.

May I ask how did you do that? Is it possible to port different deep learning framework weight?

[leondgarse]

• I'm keeping the porting script here 08-20_pytorch_to_keras.md#adaface-model.
• Ya, porting weight is not that hard. Technically, it's just loading pytorch weights as list -> align weights order matching keras model -> loading weights into keras model. You can also see in my another repo keras_cv_attention_models, that most models there are ported from pytorch, also some ported from paddlepaddle and mindspore. Most porting script also kept in my notebook above.