Deep Learning for Computer Vision

Fine-tuning several layers in deep learning

A pre-trained model can be loaded and only a few layers can be trained. This approach works better when the given problem is very different from the images that the model is trained upon. Fine-tuning is a common practice in deep learning. This gives advantages when the dataset is smaller. The optimization also can be obtained faster.

Training a deep network on a small dataset results in overfitting. This kind of overfitting can also be avoided using the fine-tuning procedure. The model trained on a bigger dataset should be also similar, as we are hoping that the activations and features are similar to the smaller dataset. You can start with the stored weights path as show below:

top_model_weights_path ...