Details about each individual application below. NOTE: results may not match perfectly to the paper but the top hypotheses and scores should be similar.
We utilize the following datasets in our applications:
- LaMeM
- ImageNetV2
- PartiPrompts
- DiffusionDB
- we use a random subset of the DiffusionDB prompts which are located in diffusionDB.csv
Note that we generate the images using the Parti and DiffusionDB prompts. These images can be generated by running generate_images.
To prepare the data CSV for each application: (1) download the dataset (2) replace the path in data_paths.py (3) run the generate_csv.py file in the subfolder of each application.