Training StackGAN++ to Generate Images with Better Quality

Follow the step-by-step process of training StackGAN++ for good-quality image generation.

We'll cover the following

The authors of StackGAN++ have open sourced the full source codehttps://github.com/hanzhanggit/StackGAN-v2.

Steps to train StackGAN++

Follow these steps to train StackGAN++ on the CUB-200-2011 dataset. Follow the following steps:

  1. Install the required libraries.

  2. Download the source code of StackGAN++https://github.com/hanzhanggit/StackGAN-v2.

⚠️ The dataset is intended only for non-commercial research and educational use.

  1. Download the CUB-200-2011 datasethttps://www.vision.caltech.edu/datasets/cub_200_2011/ and put the CUB_200_2011 folder in the data/birds directory, so that the images are located at paths such as data/birds/CUB_200_2011/images/001.Black_footed_Albatross/Black_Footed_Albatross_0001_796111.jpg. The compressed file that needs to be downloaded is about 1.1 GB in size.

  2. Download the pre-trained text embeddingshttps://drive.google.com/file/d/0B0ywwgffWnLLSW84ZXRjdXhObzQ/view?resourcekey=0-1NIY4Hno0B-vhYVrUHGfLQ and move it to data/birds. Make sure to rename the text_c10 folder to text.

Get hands-on with 1200+ tech skills courses.