Artificial intelligence (AI) trained on computer-generated images of analog clocks captured at different angles has learned to read the time.
Researchers at the U.K.'s University of Oxford enabled the system to realize 74% to 84% accurate readings on three sets of test images.
These images, plus the correct time shown by the clocks, were fed into a spatial transformer network, which can warp an image taken at an angle in order to perceive it face on.
The researchers also trained the model on time-lapse videos of clock faces, which enhanced accuracy when tested against 4,472 clock images.
Google AI's Arsha Nagrani said, "Given the challenge with getting manually labeled data, the authors creatively use synthetic data and the fact that time progresses to automatically correct pseudo labels."
From New Scientist
View Full Article
Abstracts Copyright © 2021 SmithBucklin, Washington, DC, USA