Home BlogDataset Improving the World’s Best Computer Vision Models

Improving the World’s Best Computer Vision Models

by Michael McLaughlin
by
Images of chairs laying on their side.

Researchers from MIT and IBM have released ObjectNet, a dataset of 50,000 images of objects that are intentionally placed in cluttered rooms or in unusual positions. For example, the dataset, which contains over 300 types of objects, has images of chairs placed on their side. The researchers found that placing objects in unconventional scenarios significantly reduces the performance of high-performing object detecting systems, demonstrating that object detectors can still struggle to understand that objects are three-dimensional and can exist in different contexts and in unusual positions. 

Get the data.

 

You may also like

Show Buttons
Hide Buttons