Hi everyone,
Does anyone know any tutorial to learn how to apply transfer learning or fine tuning? In my case, I want to develop a face recognition system, but many models can be used. Any tutorial that clearly explain these concepts or how to design own models?
Thanks,
Ander
Hi @andergalvan, If you want to recognize faces using transfer learning
#import a pre trained model
model=tf.keras.applications.VGG16(weights='imagenet',include_top=False,input_shape=(224,224,3))
#make the model layers non trainable so that the pre trained weights will not change during trainning
for layers in model.layers:
layers.trinable=False
now add new layers to the model which you want like dense,average pooling layers etc, but make sure that the last dense layer will have neurons equal to numbers of labels.
now you can train the model with your own data to recognize faces. Thank You.
Hi @Kiran_Sai_Ramineni
I appreciate a lot your sugestion, but I would like to learn how to choose the correct layers I need to add. Makes sense?
Thank you,
Ander
Hi @andergalvan, It’s an experimental thing. Try with different combinations of layers like convolution and dense layers and see which gives you the best results. Thank You