A popular data set used by thousands of students to build an open-source self driving car contains hundreds of instances of critical errors and omissions, wrote Roboflow cofounder Brad Dwyer in a company blog. “We did a hand-check of the 15,000 images in the widely used Udacity Dataset 2 and found problems with 4,986 (33%) of them. Amongst these were thousands of unlabeled vehicles, hundreds of unlabeled pedestrians, and dozens of unlabeled cyclists,” Dwyer wrote. Another 217 images (1.4%) were completely unlabeled but contained vehicles, street lights or pedestrians. The omission could be critically dangerous and lead to human fatalities in the real world, Dwyer said.
