Defining Optimizer Objects and single Train Step

**You have to define the Training steps and it's required objects**

The Training step should follow the following guidelines :

Inputs: (some/all of) Original Image, Mask, Image with Mask Applied, Original Content of the masked area

Steps:
- Pass input through the models
- Find Adverserial Loss (as Some linear combination of Generator loss, Global Discriminator Loss, Local Discriminator Loss)
- Apply Loss to the models using the above Optimizer Objects

No files, tests, or entry points are named. Begin by reviewing the existing model and training code, then map the listed inputs to the generator and discriminator passes and the three loss components. Done means optimizer objects and one training step implement the stated flow, including adversarial loss application.

Written by the indexing model from the issue text.