Comics_Generation in TensorFlow

NTU MLDS2017 Project 4

Project Link: https://www.csie.ntu.edu.tw/~yvchen/f106-adl/A4

Tensorflow implementation of Conditional Generative Adversarial Network (CGAN) automatically generates anime images based on given constraints (ex: green hair, blue eyes ..etc). The images are synthesized using the GAN-CLS Algorithm from the paper Generative Adversarial Text-to-Image Synthesis. The following is the model structure.

Image source is from the paper Generative Adversarial Text-to-Image Synthesis.

Dataset

• Download Link: https://drive.google.com/drive/folders/1bXXeEzARYWsvUwbW3SA0meulCR3nIhDb.
• tag_clean.csv includes multiple tags (hair color, eyes color, etc..) for each anime image.

Quick start

1. Download Dataset from link above.

2. Run the shell script!

./run.sh [test file]

• [test file] should be <testing_text.txt> in the main folder. You can modify content if want different result.

Training

• In image_generation.py, change mode = 1 (line 20) to be mode = 0, then do the Quick Start!

Demo Results

• We test our model with test file below (testing_text.txt)

1,blue hair blue eyes
2,grey hair green eyes
3,green hair red eyes
4,red hair green eyes
5,green hair blue eyes

• Generated Samples

References

• paarthneekhara/text-to-image

• carpedm20/DCGAN-tensorflow

• m516825/Conditional-GAN