The first study on computer vision suggests that neural networks are so complex that they are incomprehensible when approaching from the top to the bottom: this book looks like this> so there will be this pattern> otherwise it will look like this .
For some subjects, this is also effective, but when describing each object, from many perspectives, variations in color, motion and many other things, imagine how difficult it is.Even a baby's level of awareness will need enormous amounts of data.
The bottom-up approach mimics how the brain works seems more promising.Computers can apply transformation sequences to images and find out the contours, the objects it refers to, the angle of view, the movement, etc. This process requires a lot of computational and statistical figures, but only by number of images. the picture he used to be taught - just like the human brain.
Awareness of computer images
The image above (from Purdue University's E-lab) shows a computer that displays (according to its calculations) the highlighted objects with looks and properties as other examples of that object, according to some degree of certainty about statistics.
Proponents of this approach can say 'I told you' until recent years, the creation and operation of artificial neural networks is very difficult because the calculation is too large.Progress in parallel calculations has reduced this difficulty.The past few years have seen the explosion of research and use of this system in mimicking the human brain.The process of pattern recognition is still accelerating and we are still making progress.
Understanding
Of course you can still build a system to identify an apple, from any angle, in any situation, whether standing or moving, whether bitten or intact, but still unable to receive an orange.
It also cannot tell you what an apple is, how much it is to eat, how big or small it is to use.This means that even good hardware and software cannot do anything without an operating system.
That's the rest of the brain: short / long-term memory, sensory data, attention, awareness, lessons when interacting with the world . written on the network of connected neurons more complicated than anything we've ever seen, in a way we can't understand.
That's where computer science and artificial intelligence meet.Between computer scientists, engineers, psychologists, neuroscience and philosophers, there is still no definition of how the brain works, let alone simulate.
Although new in the early days, computer vision was still very useful.It is present in your Face ID camera and smile.It helps self-driving cars identify signs and pedestrians.It is in robots in the factory, identifying products, transmitting to humans.
The path is still long until they look like people but on that road, the things they do are amazing.
See more: