First version with the following features:
- Memory-efficient version (trade-off between memory and speed).
- Different types of upsampling (deconvolution, upsampling and pixel shuffle).
- Different types of pooling (max-pooling, avg-pooling, blur-pooling).
- The depth and width of the Tiramisu is fully configurable.
- Early-transition can be enabled when the input images are big.
- The activation function of the last layer can be disabled or modified.