TensorFlow Machine Learning Projects

DiscoGAN modeling

For each mapping, that is, handbags (denoted by b) to shoes (denoted by s), or vice versa, we add two generators. Let's say, for the mapping b to s, the first generator maps the input image from domain b to s, while the second generator reconstructs the image from domain s to domain b. Intuitively, we need a second generator to achieve the objective (one-to-one) mapping we talked about in the previous sections. Mathematically, this can be represented as follows:

While modeling, since this is a very hard constraint to satisfy, ...