# TAC-GAN - Text Conditioned Auxiliary Classifier Generative Adversarial Network

**Authors:** Ayushman Dash, John Cristian Borges Gamboa, Sheraz Ahmed, Muhammad Zeshan Afzal, Marcus Liwicki  
**arXiv:** [1703.06412](https://arxiv.org/abs/1703.06412)  
**Date:** 2017-03-19  
**Summary page:** https://www.summarizepaper.com/en/paper/1703.06412v1/

## Key points

- TAC-GAN is a state-of-the-art text-to-image GAN that uses class information to improve diversity and coherence of generated images.
- Trained on the Oxford-102 dataset of flowers, TAC-GAN outperformed previous models with an inception score of 3.45.
- The research conducted by Ayushman Dash, John Cristian Borges Gamboa, Sheraz Ahmed, Muhammad Zeshan Afzal, and Marcus Liwicki showcases TAC-GAN's effectiveness in pushing the boundaries of image synthesis technology.
- TAC-GAN demonstrated the ability to produce highly diverse images across all classes with an average MS-SSIM of 0.14.

## Summary

The Text Conditioned Auxiliary Classifier Generative Adversarial Network (TAC-GAN) is a state-of-the-art text-to-image GAN that utilizes class information to enhance the diversity and coherence of generated images. Trained on the Oxford-102 dataset of flowers and evaluated using Inception-Score and Multi-Scale Structural Similarity Index (MS-SSIM) metrics, TAC-GAN outperformed previous models with an impressive inception score of 3.45. This marks a 7.8% increase compared to StackGan and showcases the effectiveness of TAC-GAN in pushing the boundaries of image synthesis technology. The research, conducted by Ayushman Dash, John Cristian Borges Gamboa, Sheraz Ahmed, Muhammad Zeshan Afzal, and Marcus Liwicki, highlights the model's ability to produce highly diverse images across all classes with an average MS-SSIM of 0.14.