1. Learn
    2. /
  2. Courses
    3. /
  3. Deep Learning for Images with PyTorch
Connected

Exercise

Training loop

Finally, all the hard work you put into defining the model architectures and loss functions comes to fruition: it's training time! Your job is to implement and execute the GAN training loop. Note: a break statement is placed after the first batch of data to avoid a long runtime.

The two optimizers, disc_opt and gen_opt, have been initialized as Adam() optimizers. The functions to compute the losses that you defined earlier, gen_loss() and disc_loss(), are available to you. A dataloader is also prepared for you.

Recall that:

  • disc_loss()'s arguments are: gen, disc, real, cur_batch_size, z_dim.
  • gen_loss()'s arguments are: gen, disc, cur_batch_size, z_dim.

Instructions

100 XP
  • Calculate the discriminator loss using disc_loss() by passing it the generator, the discriminator, the sample of real images, current batch size, and the noise size of 16, in this order, and assign the result to d_loss.
  • Calculate gradients using d_loss.
  • Calculate the generator loss using gen_loss() by passing it the generator, the discriminator, current batch size, and the noise size of 16, in this order, and assign the result to g_loss.
  • Calculate gradients using g_loss.