metadata
license: mit
task_categories:
- text-to-image
- image-to-image
- unconditional-image-generation
tags:
- art
pretty_name: stained
size_categories:
- n<1K
Stained Glass Art Dataset for Diffusion Models
Overview
This dataset consists of 21 high-resolution images of stained glass art, accompanied by corresponding captions. It is designed for fine-tuning diffusion models using techniques such as textual inversion and dreambooth. The dataset is intended to facilitate research and experimentation in generating stained glass art-inspired images.
Dataset Structure
- Images: The stained glass art images are stored in the "images" directory, with filenames ranging from "0.jpg" to "20.jpg."
- Captions: Captions for each image are provided in the "captions.csv" file located in the dataset's root directory. The captions contain placeholders for adjectives and a custom token to represent stained glass art. For example: "A {adjective} {token} of a puppy."
- Adjective Placeholders: During training, the {adjective} placeholder in the captions can be randomly selected from the following list:
["", "good", "cropped", "clean", "bright", "cool", "nice", "small", "large", "dark", "weird"]
. - Token Placeholder: The {token} placeholder represents the custom token that needs to be trained to capture the unique art style of stained glass. This token is a key element in generating realistic stained glass art-inspired images.