A hybrid images has two coherent global image interpretations, one of which is of the low spatial frequencies, the other of high spatial frequencies. An example can be seen below:
Code for generating hybird images:
e.g., for blending image 12 to image 13 use:
python hybrid.py -i0 ./data/12.jpeg -i1 ./data/13.jpeg
Use the mass_hybrid(config) function to generate the datase (please see the code).
This code is a modified version of the original code at: https://github.com/rhthomas/hybrid-images
Dataset is available in DB.zip encompassing 10 categories (Banana, Custard Apple, Fig, Granny Smith, Jackfruit, Lemon, Orange, Pineapple, Pomegranate, and Strawberry) with 10 images per category.
If you use this code in your research, please cite this project.
@article{Borji_Hybrid_2022,
title={CNNs and Transformers Perceive Hybrid Images Similar to Humans},
author={Borji, Al},
journal={Arxiv},
year={2022}
}