SUN RGB-D is an RGB-D image dataset for scene understanding.

This dataset provides a large amount of data for training scene understanding related algorithms. The 3D metrics can be used to evaluate the algorithms and avoid overfitting to small test sets. In addition, this dataset can also be used to study cross-sensor bias.

The dataset includes:

• 10,335 indoor images of different scenes, 5,285 in the training set and 5,050 in the test set
• More than 700 object annotations
• 146,617 2D polygon annotations
• 58,657 3D borders

Example image: