Good research demands good data

"Good research demands good data" is what Dr. Su told me when I was a research assistant in Academia Sinica. In the beginning , I thought his word was sarcastic to the computer vision research, since many computer vision papers only work on the data set provided by the authors. Therefore, to be a good research assistant, I tried to make my result work on every video. Then I gave up. I realized that computer vision has its own limitation. Some techniques demand high quality video to make them work.

Recently, on David Lowe's course, he explained how to reduce specular reflection which serious affects the result of stereo matching. He said that there are some papers which propose novel algorithms to detect and reduce the specular reflection effect in images, but the most effective way is to add polarizing filter to your camera. The polarizing filter can reduce specular reflection effect to 0. Instead of producing good vision algorithms, we can put our efforts on obtaining good data. Why not?