This repository provides code corresponding to the report Logo-Transformer for Generating Fantastic Logos by Brand Names by Xiaolin Zhu and Fang Xie.

Python files include:
Logo_Transformer.py to build our proposed Logo-Transformer network;
train_logo_transformer.py to train Logo-Transformer based on the training set and visualize the training process;
test_logo_transformer.py to test model performance by presenting losses and accuracy and analysing the attention distribution;
generate_logo_transformer.py to implement logo generation inference given the required brand name and the desired style;
resize_images.py to resize logo images to different sizes;
process_logo_data_logo2k.py to cluster Logo-2K+ images into different clusters and split into training, validation and testing sets for the following experiment;
cluster.py to run process_logo_data_logo2k.py based on a pre-defined cluster number.

The dataset involved in this study can be downloaded from Famous Brand Logos and Logo-2K+.

Packages required: python(3.8.11), pytorch(2.0.0), torchvision(0.15.1), numpy, pandas, sklearn, math, matplotlib, seaborn, time, random, pickle, PIL, kneed

The file Logo_Transformer.py is based on codes of Image Transformer and Transformer for language translation.