The lecture by Dr. Terence Sim is really impressive and interesting. Due to the time limitation, the lecture mainly covers two parts: understand and create (actually it includes six parts: store, share, search, see, create, understand+ a bonus part: social); there is already so much to cover and we have learnt a great deal of interesting information.
The lecture starts with some amazing and beautiful photographs. When I knew that these lively and lifelike photographs are man-made, not be taken by a camera, I was shocked. Actually these photographs are made with computer photographic techniques. (I do not understand how to make it in details now) According to my notes and my memory, these photographs are “scanned” by the reflection and then the drawer plots it in exact details like chiaroscuro on the device.
Then we come to the part of good ideas. The first good idea is models. According to the definition, models are the approximations of reality. The reason we need models is because the reality is often complex. For example, when applying Einstein's principle of relativity, we have to reach some demanding conditions, like reaching the speed of light which are very difficult to reach in real life so that we have to apply modals to reach these ideal conditions to solve some problems. One application is Ray tracing. The photographs shown in the beginning are made by the technique: Ray tracing. However, the process is painful, a photograph like a pile of cobbles needs 4.5 days to complete. Because of the long, painful process, it is mostly used in movies.
The second good idea is collecting lots of data. The data includes different scenes, light and color which are filmed by lots of cameras. Since there are many cameras, we can get data not only from these existed cameras, but also from these invisible, increased cameras. After we record and recombine all the data, we will get different views and further, the 3-D images. This technique is used in many different fields, like concerts and competitions.
The third good idea is bringing more knowledge to bear. For example, there is photograph where a man wearing a pair of sunglasses. If we want to change this photograph into a man without the sunglasses, we do not need to take another photograph; instead we just apply corresponding technique (statistical correlation between parts) to make the changes.
The forth good idea is machine learning. One important application is face detection in cameras. As far as I understand, it has a function which divides an image into tiny regions and then compares the tiny image with the format of people’s face to find the face. (Please forgive me my possible mistakes)
Then we come to the bad idea parts. However, when covering this part, the time is running out so that we quickly viewed through the slides and just covered some parts. The first bad idea is imitating human vision. At first, I was a little confused, because it is human’s imagination that promotes the development. But after the explanation, for example, the history of flying, I understand that it is because human vision is not well understood. People are eager to fly in the sky for thousands of years and they thought the reason why birds could fly was the wings. So many persons tried to fly with wings for a long time with the end of failure or even death. This example shows that the purposes of human’s desire are not bad, but if we cannot handle the principles and apply mechanisms, we will just end with failure.
The next part is motivating example. The purpose is to change a source image into the target image based on an existed image. To simplify this problem, we ignore the color first. Then we make a graph of plot count vs. pixel value (we assume the images are the same size). We manipulate statistics to match the mean and standard deviation. After a long process of calculation and applying lots of complex formulae, we finally get our white and black, target image. It is much more complex to change the color with more formulae and longer process of calculation. And the most amazing point is that we can use Photoshop to finish all the previous complex procedure within a minute which actually consists of all the procedure.
The other bad idea is that do not push math too far and there are many slides about this. (But we have no time to go through it) And the final part is the bonus part, social. We mainly cover one example: Google Street. It is not hard to figure out that there is something outside the engineering field which we have to pay attention to. Therefore, when applying some new technique, we have to consider more about people’s needs.
No comments:
Post a Comment