This dataset contains the embeddings of each image in ImageNet. The embeddings were produced by the OpenAI implementation of CLIP with the ViT-L/14 variant using the snapshot at
https://openaipublic.azureedge.net/clip/models/b8cca3fd41ae0c99ba7e8951adf17d267cdb84cd88be6f7c2e0eca1737a03836/ViT-L-14.pt.
The version of ImageNet used was the full set in the Winter 2021 release, MD5: ab313ce03179fd803a401b02c651c0a2.
There are 13,158,856 images in the dataset. The file index.parquet has two… See the full description on the dataset page:
https://huggingface.co/datasets/kinianlo/imagenet_embeddings.
