This toy dataset contains 5 images for fine-tuning text-to-image model with textual inversion approach.
This version of a dataset does not contain any captions as they're taken from a dataset implementation according to the TI paper.
- Number of images: 5
- Image resolution: 1024x1024
The images were arbitrarily taken on phone camera and cropped afterwards.
Since default captions themselves may not quite accurately describe the style or object presented, TI embeddings tend to overfit. Solution: give more meainigful and precise captions.
The dataset is provided in a following format:
images/directory