I used the Danbooru Full-body Anime Dataset from Kaggle for this project.
- Source: https://www.kaggle.com/
- Total images in dataset: around 337k
- Images I actually used: 1,024 for training, 257 for validation
- How I selected them: random sampling, no special filtering
I needed colorful anime-style images to train a manga colorization model.
Danbooru is a large public anime image board, and someone already compiled it into a Kaggle dataset, so it was convenient to use for coursework.
Copyright
- The dataset is publicly available on Kaggle
- I only use it for university coursework
- I’m not selling the model or using it commercially
- Outputs are just for demonstrating how GANs work
Privacy
- All images are anime drawings, not real people
- No personal information or identifiable data
- No photos, documents, or real identities involved
Bias
- The dataset is mostly Japanese anime style
- The model probably won’t work well on photos or western comics
- Results are limited to anime/manga-style artwork
Content
- I tried to avoid weird or inappropriate images
- Mostly normal anime character illustrations
- No realistic or sensitive content
Just this coursework.
Learning how image-to-image GANs work.
No plan to release the model publicly or use it outside the assignment.
- Dataset is from a public source
- Only used for educational purposes
- No personal data involved
- Thought about possible biases
Name: Skye Chiu Zhao
Date: 12/26/2025