One of the many interesting things that were shown was povray.org. The rendered pictures on the site looks exactly like real photos taken. It was amazing. The details were immaculate. On first look, many were unable to tell whether it was rendered or not. It was "stunning". Thereafter, Dr Sim brought up the Graphics Turing Test, which I found to be very, very interesting. Modern day computer graphics are so advanced nowadays that it might be possible that a human is unable to tell whether it's a rendered image or a real image.
"The Graphics Turing Test is a variant of the Turing Test, the twist being that a human judge viewing and interacting with an artificially generated world should be unable to reliably distinguish it from reality." - Wikipedia
On to the points that were covered by Dr Terence Sim. He touched on the good points on how visual computing can mimic real life. Firstly, he brought up Models which are essentially the approximation of reality. One good example is Ray Tracing whereby an image is generated by tracing the path of light through pixels in an image plane. This is thus capable to produce very realistic images. However, it takes up too much time. Even with rendering farms to render the image in parallel, it will still take days to fully render the picture. Hence, this technique is more often used for movies than games. Another example is a physics-based simulation whereby physics models are used to sculpt the image. This can be done for waves etc.
Secondly, the computer can collect a lot of data to produce high degrees of photorealism. There are techniques like using many cameras to take many pictures per second. This will tell how the object moves and proceed to view scenes from other viewpoints. There is also a possibility of creating virtual images/videos with this data.
Thirdly, it is to bring more knowledge to bear. The user can expose the computer to many similar objects from other sources. After which, the computer can exploit these images. One example which was shown to us was the combination of visible image and near infra-red image to form a sharper picture. There is also computational photography whereby the computer will take pixels from other sources and fill it in correctly into the new picture due to the similarities that were shown.
Another point is machine learning which I find amazing. Certain things are unable to be described fully such as faces. Only by showing various pictures of faces to the computer can we let the machine "learn" to figure out what is a face. After which, there was algorithms to tell the machine to differentiate between positive examples and negative examples. Hence, comes face detection. This can also be known as object-class detection. A video was also shown that a vehicle can drive by itself using cameras and algorithms to detect. There are certain cars that have driver-assist systems installed such as the Lexus and the BMW. This technology might find its way into daily cars soon.
Lastly, he touched on a bad point for visual computing - which is to imitate human vision. Most of the time the result is unsuccessful. Take for example the early days of flying, when people thought flapping wings were the way to fly. Human vision is still very vast and an unknown principle as of yet, thus we should not base visual computing on it. Also, there's a point on illusions which we still find it a mystery.
Well, I must say I enjoyed the seminar a lot, visually. Thanks Dr Terence Sim for sharing what you like! Simply stunning!
No comments:
Post a Comment