Add pytorch_Pix2Pix_cGAN.py: implementation of Pix2Pix with conditional GAN (cGAN)

[anh-nn01]

Primary contributions:

1. Add implementation for Pix2Pix using cGAN, a general framework for image translation tasks (e.g: Satellite to Maps, Image to Anime, Scene Day to Scene Night, etc.). Reference paper: https://arxiv.org/abs/1611.07004.
2. Add implementation of U-Net model for Pix2Pix's Generator.
3. Add implementation of CNN Binary Classifier for Pix2Pix Discriminator.
4. Add implementation of the cGAN training framework consisting of U-Net Generator and CNN Discriminator.
5. Modified checkpoint saving mechanism: saving checkpoint every 5 epochs.
6. Provided hyperparameters for training Pix2Pix, specifically batch_size, learning_rate, num_epochs, beta_1 and beta_2 (in Adam Optimizer), L1 regularizer, and number of GPUs.

IMPORTANT NOTE (for README.md):

1. To test the new Pix2Pix framework, we test the model on the Satellite Imagery to Map the translation task.
2. The task dataset should be downloaded on the same level of pytorch_Pix2Pix_cGAN.py.
3. The dataset can be downloaded here .
4. If you want to try the model on another task, such as Image to Sketch, please download the dataset and change the dataset directory in pytorch_Pix2Pix_cGAN.py.

[anh-nn01]

Please ignore this file in this commit and only care about requirements.txt. The correct pytorch_Pix2Pix_cGAN is in the later commit.