- About the Corel-Princeton Image Similarity Benchmark
- This benchmark was created based on a human subject
study of 121 people.
- For each query image, the groundtruth contains not
only the similar images, but also their similarity scores
assigned by our human subjects.
- Image dataset:
- This dataset contains 10,000 images randomly picked
from the Corel Stock Photo Library.
- dataset.txt lists the
10,000 images' ids in the original Corel dataset, in the
order that they were picked.
- Query images:
- Eight query images are used in this benchmark: airplane,
beach, car, flower, horse, fruit, model, columns.
- queries.txt lists the query
images and their corel ids.