Which Vision Succeeded?

It is interesting to note that certain experiments were carried out in Germany to address the void between computer and human vision.

Experiment 1 was conducted to understand how neural networks perceive contours (or boundary). For this, both the human and the AI participant were shown an image and asked whether it has closed contour or not. The experiment was specifically done to check if deep learning algorithm can identify closed and open shapes or not.

Results showed that a well-trained neural network could identify a closed contour even when the contour was not made of straight lines.

Experiment 2 was performed to understand whether the deep learning algorithm could identify the relation between different shapes in the picture or not. While human vision could easily understand this, it was found that deep learning algorithms that were trained with less found it difficult to draw a conclusion.

Experiment 3 was specifically carried out to identify the recognition gap that strains computer vision technology to detect an object or a thing from closer proximity when it is zoomed. When we zoom in an image beyond a certain range, we are less likely to understand what it is. However, deep learning performed better than humans and could identify minuscule features that are imperceptible to the human eye and can be detected only when the image is zoomed in very closely.