This object-recognition dataset stumped the world’s best computer vision models

MIT News  December 10, 2019
In the real-world object detectors’ performance drops noticeably creating reliability concerns for self-driving cars and other safety-critical systems that use machine vision. A team of researchers (MIT, IBM) created ObjectNet consisting of about 50,000 photos of objects shown tipped on their side, shot at odd angles, and displayed in clutter-strewn rooms and it contains no training images. When leading object-detection models were tested on ObjectNet, their accuracy rates fell from a high of 97 percent on ImageNet to just 50-55 percent. The researchers hope that the new dataset will result in robust computer vision without surprising failures in the real world. They will present their work at the Conference on Neural Information Processing Systems (NeurIPS)…read more.

ObjectNet, a dataset of photos created by MIT and IBM researchers, shows objects from odd angles, in multiple orientations, and against varied backgrounds to better represent the complexity of 3D objects. Credit: MIT

Posted in AI and tagged , .

Leave a Reply