seonglae/vls-10k

VLS 10K 9,987 MS-COCO images paired with a long written description, a one-sentence spoken summary of that description, the spoken audio, and that audio pre-encoded by two neural codecs. Images and audio are embedded in the parquet, so the viewer renders them and one call opens the set: from datasets import load_dataset ds = load_dataset("seonglae/vls-10k", split="train") ds[0]["image"] # PIL image ds[0]["audio"] # decoded waveform ds[0]["sst"] # the sentence that was… See the full description on the dataset page: https://huggingface.co/datasets/seonglae/vls-10k.