This project is our own implementataion of text-to-image generation for birds. Based off of descirpiton provided by the user, it tries to create an original bird image. It runs in Python 3 and uses a target-aware generative averserial model.
youtube.com/watch?v=FgwgQJRBPAc
https://colab.research.google.com/drive/1mfxWs4v8WekAs4snLyXvPenQz0g7b2ZY?usp=sharing
After cloning the github, change directory into it. If using anaconda, run command
conda env create environment.yml
conda acivate text-to-image-generation-env
into the console to activate the yaml file with all the required libraries. Additionally, install gdown in the comand prompt after activating the environment
birds
Contains dataframes for filenames (filename.pickle) and info of classes (class_info.pickle) for each image for testing and training.
logs/train
Logs training progress over time.
environment_check.py
checks that the right versions of libraries from the environment were installed.
environment.yml
File use to install the required libraries.
test
Tests used during development.
requirements.txt
Required packages for code.
stage1_dis.h5
Checkpoint for discriminator after training.
stage1_gen.h5
Checkpoint for generator after training.
test-requirements.txt
Requirements for testing in git workflow
To begin training with the dataset we used, run python model.py. It checks if the CUB_200_2011 directory that contains the dataset exists, and if not, downloads and sets up the directory for it. The model trains the data, and saves them in the current directory.
Han Zhang, Tao Xu, Hongsheng Li, Shaoting Zhang, Xiaogang Wang, Xiaolei Huang, and Dimitrius Metaxas. StackGAN: Text to Photo-realistic Image Synthesis with Stacked Generative Adversarial Networks," arXiv, August 5, 2017. [Online]. Available: https://arxiv.org/abs/1612.03242v2. [Accessed December 06, 2021]