I’m Yuma Ouchi, M1 student. I have finished internship for one month at University of Oulu, so I would like to make a final report on it.

As I wrote in the interim report, I learned the basis of multi-view stereo in this internship. The reason for setting this theme is that it is an important task to generate 3D models from 2D images of human in my research.

The above images are 3D point cloud generated using multi-view stereo algorithm. The left is obtained from 16 input images, and the right is obtained from 312 input images. From this result, we can see that 3D reconstruction with multi-view stereo requires many input images.

I would like to realize 3D reconstruction using as few cameras as possible in a large space. Therefore, it turned out that it is insufficient only with multi-viewpoint stereo approach. Therefore, it turned out that the current multi-viewpoint stereo approach alone is insufficient. Ultimately, Prof.Janne proposed a method using machine learning, so the direction that I should address in the future was clarified.

Next, I would like to mention Finland.  The temperature will be less than -10 ° C under the influence of radiation cooling phenomenon on sunny days. At that time it was the limit to walk outside of 10 minutes. However, life in Oulu was very comfortable. There was a shopping mall in one minute on foot from the hotel, and some stores were open late at night. And, above all, the cityscape is beautiful.

Although it was a short period, I had a valuable experience. I would like to engage in research by taking advantage of experience in Oulu.