HD View was developed in my group at Microsoft Research. I very much enjoy working on research projects like HD View. The end goal is a demo application that we can share with the world, but along the way we sometimes develop novel solutions to the problems that we encounter. When we develop something new we almost always look for a suitable journal or conference to publish it. The publication process can be humbling. It involves peer review by experts in the field and what you think is a great idea doesn’t always pass muster with the reviewers. As part of the HD View work, we submitted 2 papers to Siggraph, the premiere conference in computer graphics. A few weeks ago we were very pleased to find out that they both successfully passed through the peer review process and were accepted.
The first paper is titled Joint Bilateral Upsample. This paper was motivated by the very large pictures that we were trying to process. Today’s desktop computers aren’t powerful enough to apply a state-of-the-art tone mapper to a 4Gpixel image. Instead we applied the tone mapper to a smaller version of the image and used the full resolution image to intelligently guide the process of upsampling the tone mapping parameters. We then realized that this same technique could be applied to other problems as well.
The second paper is titled Capturing and Viewing Gigapixel Images. This paper describes the capture and viewing process that we developed. There are some interesting details in the capturing section, but the part I find most interesting is the work we did on viewing big/wide/deep images. The two main contributions in viewing are dynamically adjusting the projection of a wide field of view image and dynamically adjusting the tone of the image given the current view. The first of these you can see in HD View today. As we zoom in, we apply a perspective (also known as rectilinear) warp as we zoom out we apply a spherical (also known as equirectangular) warp. In between, we developed a new technique to smoothly transition from one to the other. The tone adjustment will be released in HD View sometime soon. The motivation behind this is that these massive images almost always have very deep dynamic range. The tone adjustment of the image varies based on the particular part of the image that is currently being viewed. This isn’t something that can be computed once, when the image is created, rather it has to be done on the fly as the user interacts with the image.
That’s probably enough details for a blog entry. There are many more details on the project pages that Johannes Kopf has put together for Joint Bilateral Upsample and Capturing and Viewing Gigapixel Images. Included there are the papers, videos and soon some source code.