TensorFlow in Action

7 Teaching machines to see better: Improving CNNs and making them confess

This chapter covers

• Reducing overfitting of image classifiers
• Boosting model performance via better model architectures
• Image classification using pretrained models and transfer learning
• Modern ML explainability techniques to dissect image classifiers

We have developed and trained a state-of-the-art image classifier known as the Inception net v1 on an object classification data set. Inception net v1 is a well-recognized image classification model in computer vision. You learned how Inception blocks are created by aggregating convolution windows at multiple scales, which encourages sparsity in the model. You further saw how 1 × 1 convolutions are employed to keep the dimensionality ...