How can I use importKerasNetwork function to Import a pretrained VGGFace Keras network and weights?

9 次查看(过去 30 天)
How can I use importKerasNetwork function to Import a pretrained VGGFace Keras network and weights and use it for transfer learning?

采纳的回答

Maria Duarte Rosa
Hi Ibrahim,
layers = importKerasLayers(modelfile) imports the architecture and weights of Keras models saved as .h5 files. These files have to be generated from Keras.
After importing the layers, these can be modified for transfer learning as described here:
I hope this helps,
Maria
  2 个评论
Ibrahim Mohammed
Ibrahim Mohammed 2019-7-6
Thanks Maria for the quick response,
I couldn't find the keras version of VGGFace in the official keras library, but I found one on github at this link: https://github.com/rcmalli/keras-vggface. Is there any way to access this model using importKerasLayers function so that I can use it for transfer learning?
Thanks.

请先登录,再进行评论。

更多回答(1 个)

Maria Duarte Rosa
Hi Ibrahim, you first need to install and run the python package as described in the GitHub page to obtain the model (including weights), then save this model as .h5 file. Once you have the model saved in .h5 you can use importKerasLayers to load the model into MATLAB.
  4 个评论
Alex Zai
Alex Zai 2020-12-2
Hello dears,
i have a few question. I want to use pretrained facenet model in guthub site, in the following link: https://github.com/nyoki-mtl/keras-facenet.
In this link already had trained model (facenet.h5) file. How can i import to Matlab it, please?
I tried to used any suggestion like that : importKerasNetwork, importKerasLayers and downloaded and saved in matlab location, then tried read (h5info). But i can't imported which i wanted that (facenet_keras.h5) to Matlab.
If any suggestion or someone can help me, please...
Thank you very much.

请先登录,再进行评论。

类别

Help CenterFile Exchange 中查找有关 Deep Learning Toolbox 的更多信息

产品


版本

R2018b

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by