Exercises

The following exercises are of fundamental importance and you are invited to answer to every theoretical question and solve all of the code challenges presented:

1. What is the semantic segmentation?
2. Why is semantic segmentation a difficult problem?
3. What is deconvolution? Is the deconvolution operation in deep learning a real deconvolution operation?
4. It is possible to use Keras models as layers?
5. Is it possible to use a single Keras Sequential model to implement a model architecture with skip connections?
6. Describe the original U-Net architecture: what are the differences between the custom implementation presented in this chapter and the original one?
7. Implement, using Keras, the original U-Net architecture.
8. What is a DatasetBuilder?
9. Describe the hierarchical organization of TensorFlow Datasets.
10. The _info method contains the description of every single example of the dataset...