This is the dataset we collected for our Multi-Concepts learning research, if you found this dataset useful, plese cite our work.
An Image is Worth Multiple Words: Discovering Object Level Concepts using Multi-Concept Prompt Learning
We generate and collected a Multi-Concept-Dataset including a total of ~ 1400 images and masked objects/concepts as follows
/ (370 images)
/natural_2_concepts
/natural_345_concepts
/real_natural_concepts
| Data file name | Size | # of images |
|---|---|---|
| medical_2_concepts | 2.5M | 370 |
| natural_2_concepts | 36M | 415 |
| natural_345_concepts | 13M | 525 |
| real_natural_concepts | 5.6M | 137 |
If you make use of our work, please cite our paper:
@article{jin2023image,
title={An image is worth multiple words: Learning object level concepts using multi-concept prompt learning},
author={Jin, Chen and Tanno, Ryutaro and Saseendran, Amrutha and Diethe, Tom and Teare, Philip},
journal={arXiv preprint arXiv:2310.12274},
year={2023}
}