YFCC15M

YFCC15M Recaption Dataset This YFCC15M Dataset is filtered by DeCLIP and recaptioned utilize the diverse description generation framework proposed in RWKV-CLIP. The text is a list of text tokens with a length of 77, encoded using the CLIP tokenizer. You can use from clip.simple tokenizer import SimpleTokenizer as Tokenizer to decode it back into the original text. Using Dataset You can easily download and use the arxiver dataset with Hugging Face’s datasets library.… See the full description on the dataset page:

Source: Hugging Face Hub (Kaichengalex/YFCC15M). Metadata imported from the dataset’s Hub tags.