摘要: Computer Vision models have learned to identify objects in photos such precision that some may outperform humans in some data sets. But when these same object detectors are released into the real world, their performance decreases dramatically, creating reliability problems for self-driving cars and other critical systems for security that using the machine vision .
Computer Vision models have learned to identify objects in photos such precision that some may outperform humans in some data sets. But when these same object detectors are released into the real world, their performance decreases dramatically, creating reliability problems for self-driving cars and other critical systems for security that using the machine vision .
In an effort to bridge this performance gap, a team of MIT researchers and IBM set out to create a very different kind of game data object recognition. Its called ObjectNet, a play on IMAGEnet, the database photo crowdsourcing responsible for launching a large part of modern artificial intelligence development. Unlike IMAGEnet, featuring photos taken from Flickr and other social media sites, ObjectNet features photos taken by freelancers paid. The objects are presented at the end of their side, shot at odd angles, and displayed in the halls littered clutter. When main detection object models were tested on ObjectNet, their accuracy rate reached a peak of 97 percent on IMAGEnet only 50-55 percent.
......
So ObjectNet is harder than ImageNet...https://t.co/KeY837Njoc
— Subhodip Biswas (@dataquidnunc) December 11, 2019
Full Text: MIT NEWS
若喜歡本文,請關注我們的臉書 Please Like our Facebook Page: Big Data In Finance
留下你的回應
以訪客張貼回應