Labelme Image Dataset

* This dataset supports online use.Click here to jump.

Labelme Dataset is an image dataset for target recognition, covering more than 1,000 fully annotated and 2,000 partially annotated images. The partially annotated images can be used to train labeling algorithms. The test set has images taken from different parts of the world, which ensures that there will be large differences between the images in the continuation and the test.

This dataset was released by MIT – Computer Science and Artificial Intelligence Laboratory in 2007. The related paper is “LabelMe: a database and web-based tool for image annotation”, and the project has related challenges.