- Sketch-A-Net was able to distinguish between sketches of seagulls, pigeons, and other flying birds with 42.5 percent accuracy. The human subjects tested only managed 24.8 percent.
- The software uses deep neural networks to identify drawings. It was trained on the TU-Berlin dataset, which contains 250 categories of sketches with 80 drawings per category.
- Sketch-A-Net will be presented by researchers from the Queen Mary University of London at the British Machine Vision Conference later this year.
You can read the full paper on arxiv.org.