search

yinguobing / cnn-facial-landmark

Training code for facial landmark detection based on deep convolutional neural network.
MIT License
623 stars 183 forks source link

Using the pre trained model to train my own dataset ( transfer learning ) #64

Open abdou31 opened 4 years ago

abdou31 commented 4 years ago

I have trained my own dataset and I have created my own model from scratch but it doesn't give good results. The problem here that eye region landmarks = 40 points and for pre-trained model 68 points ( facial region ). I think that is possible to train my own dataset from the pre-trained model that you give it using "Transfer learning", but I don't have any idea how to use this technique on the pre-trained model that you give. Do you have any idea how can I use that?

yinguobing commented 4 years ago

TensorFlow's official model shows how to do transfer learning here: https://github.com/tensorflow/models/blob/master/official/resnet/resnet_run_loop.py

Pay attention to these keywords: pretrained_model_checkpoint_path and fine_tune.

abdou31 commented 4 years ago

Thanks. This is really so complicated, but I will try to use this script. Can you tell me what function should I use?

abdou31 commented 4 years ago

What do you think in my case for fine_tune? Should I make false or change it to true?

yinguobing commented 4 years ago

I think it should be set to True but you need to figure this out through reading the code. It's been a while since the last time I read it.

abdou31 commented 4 years ago

Thanks. I will see it.