Nouns
Nouns is a multimodal dataset composed of images and text legends automatically generated from visual attributes, intended for training image generation models.
Description
Nouns contains nearly 50,000 JPEG images with text descriptions that are automatically generated based on attributes, colors, and objects in the images. This dataset is primarily used to train models for generating images from text.
What is this dataset for?
- Train text-to-image models to generate images from descriptions
- Studies on the relationship between visual attributes and natural language
- Evaluation and improvement of automatic captioning models
Can it be enriched or improved?
Automatically generated legends can be refined or completed manually to improve the quality of the annotations and the relevance of the descriptions for training.
🔎 In summary
🧠 Recommended for
- Computer vision researchers
- Generative AI developers
- Captioning projects
🔧 Compatible tools
- Hugging Face Datasets
- PyTorch
- TensorFlow
- Diffusers
- Stable Diffusion
💡 Tip
Complete legends for specific tasks or to improve model accuracy.
Frequently Asked Questions
What is the nature of the legends in this dataset?
Captions are automatically generated based on visual attributes, colors, and objects in the images.
How many examples does the dataset contain?
Approximately 49,859 images with their associated text descriptions.
Is it a dataset suitable for beginners?
Yes, it's easy to use and well-structured, ideal for early text-to-image projects.
