Resumen de CGI humans help computers understand us
At the moment, the best computer vision algorithms are trained using thousands of images that have been painstakingly labeled to highlight key features. This is how they learn to distinguish an eye from an arm. Ideally, every pixel in every frame of a video would be labeled. "But this would mean instead of creating thousands of annotations, people would have to label millions of things, and that's just not possible," says Gul Varol at the Ecole Normale Superieure in Paris. So Varol and his colleagues have made thousands of videos of "synthetic humans." They walk, they run, they crouch--and since the clips are computer-generated, every frame is automatically labeled with all the important information. In total, the team generated more than 65,000 clips and 63 million frames. Randomly selected body shapes and clothing were used for the figures, each one set in different poses
