jacopoteneggi/IBYDMT

Pre-computed vision-language model image embeddings

Embeddings are stored as Parquet files with the following structure:

<DATASET_NAME>_<OP>_<MODEL_NAME>.parquet

"""
    DATASET_NAME:   name of the dataset, e.g. "imagenette".
    OP:             split of the dataset (either "train" or "test").
    MODEL_NAME:     name of the model, e.g. "clip_vit-l_14".
"""

dataset["embedding"]    contains the embeddings
dataset["label"]        contains the labels

To generate the dataset, run

$ python make_dataset.py

Supported dataset names (see supported_datasets.txt):

• imagenette [dataset]

Supported model names (see supported_models.txt):

• clip:ViT-RN:50 [model]
• clip:ViT-B/32 [model]
• clip:ViT-L/14 [model]
• open_clip:ViT-B-32 [model]
• open_clip:ViT-L-14 [model]
• FLAVA [model]
• ALIGN [model]
• BLIP [model]

References

@inproceedings{teneggi24testing,
  title={Testing Semantic Importance via Betting},
  author={Teneggi, Jacopo and Sulam, Jeremias},
  booktitle={The Thirty-eighth Annual Conference on Neural Information Processing Systems},
  year={2024},
}