Computer vision can be a very powerful tool for making computer vision algorithms more effective, but not all the time.
In a recent paper, computer vision researchers from Oxford University and the University of Edinburgh asked whether using computer vision for computer vision was better than using paper to create computer vision models.
They compared the results of computer vision systems that used computer vision and natural language processing to create human-like visual representations.
The result?
The researchers found that the two methods worked equally well, but that the paper was more accurate in creating visual models.
“We found that using a paper-based model is better than the computer-based method in predicting human-based human facial expressions,” lead author David Wills, a PhD student at Oxford, said in a press release.
“However, using computer-generated models to create the human-looking face is better in predicting a human-level facial expression, which is an important goal in the field of computer-assisted translation.”
While there’s no guarantee that a computer-vision system will perform better in translating a human face into a computer model, the researchers found a significant advantage in using computer images to create a computer vision model.
“Our results suggest that the advantages of using computer graphics over natural language for visual representation of a human are limited,” the researchers wrote.
“In order to overcome these limitations, we propose to use computer- and human-vision models together for human-to-computer translation.”
This article was originally published on Science Daily.
