LVIS Vocabulary Instance Segmentation Dataset

LVIS stands for Large Vocabulary Instance Segmentation, which is a dataset for long-tail instance segmentation. The dataset annotates more than 1,000 target categories in 164k images and provides 2 million high-quality instance segmentation masks.

Due to the Zipf’s law distribution of natural image categories, LVIS naturally has long-tail categories even with few training samples, which is of great significance for low-sample learning.