add readme
Browse files- README.md +44 -0
- anybooru_grid.png +3 -0
README.md
ADDED
|
@@ -0,0 +1,44 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: openrail
|
| 3 |
+
---
|
| 4 |
+
|
| 5 |
+
# Anybooru
|
| 6 |
+
|
| 7 |
+
## Synthetic Anime Image Dataset
|
| 8 |
+
|
| 9 |
+
Synthetic anime image dataset generated using Andite's Anything-v4.5 checkpoint with Danbooru2021 tags collected in https://gwern.net/danbooru2021.
|
| 10 |
+
See https://github.com/1lint/anybooru for details and code to generate your own variant of the dataset. I have also uploaded the extracted Danbooru2021 tags to https://huggingface.co/datasets/lint/danbooru_tags, this dataset was generated with a small subset of the tags in `2021_0_pruned.parquet`.
|
| 11 |
+
|
| 12 |
+
Each string of tags was used to generate 4 different images with different seeds. This serves a similar purpose as random resize crop and image flip transformations to train the model to focus on general concepts encoded in the tags, rather than memorizing specific images.
|
| 13 |
+
|
| 14 |
+
|
| 15 |
+
## Quick Start
|
| 16 |
+
|
| 17 |
+
```
|
| 18 |
+
from datasets import load_dataset
|
| 19 |
+
|
| 20 |
+
dataset = load_dataset('lint/anybooru')
|
| 21 |
+
|
| 22 |
+
sample = dataset['train'][0]
|
| 23 |
+
|
| 24 |
+
image = sample['image']
|
| 25 |
+
tags = image.info['tags']
|
| 26 |
+
|
| 27 |
+
print(tags)
|
| 28 |
+
```
|
| 29 |
+
|
| 30 |
+
## Samples
|
| 31 |
+
|
| 32 |
+
Each row of samples share the same generation prompt (string of tags).
|
| 33 |
+
|
| 34 |
+
<p align="center"><img src="./anybooru_grid.png" alt= “Anybooru” width="50%" height="50%" align="center"></p>
|
| 35 |
+
|
| 36 |
+
|
| 37 |
+
|
| 38 |
+
## Citations
|
| 39 |
+
|
| 40 |
+
```
|
| 41 |
+
@misc{danbooru2021, author = {Anonymous and Danbooru community and Gwern Branwen}, title = {Danbooru2021: A Large-Scale Crowdsourced and Tagged Anime Illustration Dataset}, howpublished = {\url{https://gwern.net/danbooru2021}}, url = {https://gwern.net/danbooru2021}, type = {dataset}, year = {2022}, month = {January}, timestamp = {2022-01-21}, note = {Accessed: 03/01/2023} }
|
| 42 |
+
```
|
| 43 |
+
|
| 44 |
+
|
anybooru_grid.png
ADDED
|
Git LFS Details
|