What Is Transfer Learning (Feature Extraction) in TensorFlow?

Previously, I have described a simple Convolutional Neural Network, which classified bird species with only 50% accuracy. The network architecture was similar to Tiny VGG and had too many parameters leading to overfitting. Image classification is a complex task. However, we can approach the problem while reusing state-of-the-art pre-trained models. Transfer learning is a machine learning technique that reuses patterns learned by a model on one dataset to improve performance on a different, related task. This way, we can efficiently apply well-tested models, potentially leading to excellent performance.

In this post, we will focus on Feature Extraction, one of the Transfer Learning techniques. I will build on the code and ideas previously shared in my previous post “Convolutional Neural Networks for Image Classification.” We will reuse previously created feature extraction models available at the TensorFlow Hub for our task of bird species recognition using image data from Kaggle. At the end of this post, we will see how this approach will improve our bird species prediction model accuracy of 50% to over 90%.

Downloading the 400 Bird Species Dataset from Kaggle

🔒 Subscribe to keep reading.

Visualizing the 400 Bird Species Image Data

🔒 Subscribe to keep reading.

Preprocessing and Augmenting Bird Images with ImageDataGenerator

🔒 Subscribe to keep reading.

Feature Extraction with Pre-trained TensorFlow Hub Models

🔒 Subscribe to keep reading.

Predicting Bird Species with the Trained Models

🔒 Subscribe to keep reading.

Saving and Loading a Trained Keras Model

🔒 Subscribe to keep reading.

Key Takeaways: Feature Extraction with TensorFlow Hub

🔒 Subscribe to keep reading.

References

🔒 Subscribe to keep reading.