Looking for HDR images of dynamic scenes for SIGGRAPH paper

[pradeepsen]

Hi everyone,

We are working on a new algorithm for producing HDR images that we hope to submit to SIGGRAPH, and so we are looking for beautiful data sets to test our algorithm and to publish in the final paper.

Specifically, we are working on a new technique that can automatically register bracketed exposures of DYNAMIC content during the HDR merging process. Most of the time when taking HDR pictures, one tries to take pictures of STATIC scenes with a tripod, because registration of dynamic scenes is difficult and one gets ghosting when merging the final HDR image if the subjects are moving too much from frame to frame. We think we have a way of merging beautiful HDR images from a set of varying exposures of dynamic content without any manual intervention.


Last year, we had a paper at SIGGRAPH that introduced a new camera that could produce HDR video of dynamic scenes:
A Versatile HDR Video Production System

We are hoping to eventually do the same thing, but without the need of a special camera. So if you have some interesting HDR shots that you think would make a great image in a SIGGRAPH paper please send them to us.

Technical requirements:

1. The data sets should be a set of bracketed exposures of interesting dynamic content taken over time. SIGGRAPH is obviously looking for very interesting and striking images, so please send us image sets that you think would produce a beautiful HDR image if one could merge them. You can send us either RAW or JPEG files, but RAW files are obviously preferred. Your sequence of exposures can have any number of images (from 2 on up), but we expect that the best results will typically be with 3 - 9 exposures, perhaps 1 - 2 stops apart.

2. The images can be handheld (with camera motion from frame to frame) and should be of dynamic content that is impossible to merge with traditional techniques without a lot of manual intervention. It would be best if the dynamic motion is fairly complex so that traditional algorithms to try to register them (such as optical flow) would not work. For example, a moving car that is simply translating across the image might be easy to register with optical flow, but people walking or pets moving in complex ways would not. Other things that would make good test scenes are dynamic objects that don't have well-defined structure like spraying water, fire, fountains, etc.

3. When bracketing your exposures, you can change both the shutter time and the aperture (either one or both). We are experimenting with various test sets and believe we can correct depth-of-field problems that would occur when varying the aperture. Of course, if you have traditional change-the-shutter-time-only bracketed exposures that might work as well (although sometimes dynamic content might become overblurred in the longer exposures, which might be difficult to fix).

4. We would like to especially see pictures where the dynamic objects themselves are of high-dynamic range. One hack that is often used for HDR imaging of dynamic scenes is to erase dynamic objects in all frames except one. This produces images that look like HDR, but where the dynamic objects are really only LDR. We would like to see scenes where the dynamic object needs to be in HDR for the final image to look good.

5. We would like the dynamic range of the scene to be as large as possible. This would preclude another common hack which is simply to exagerate the tonemapping of a single RAW image. We want the scenes to have a large dynamic range so that doing this simple trick would not produce acceptable results.

6. We need the information of your camera that you used to take the pictures, as well as the camera exposure settings so that we can merge the final HDR image.

7. You should have taken these pictures yourselves and have the legal rights to them. We will ask you to fill out a permission slip that allows us to use your images in our technical paper.

If we end up using your image(s) in our paper, we will send you the final HDR generated from your images and we will also acknowledge you as the photographer in the final version of the paper. Please understand that when we post a request like this on a forum, we often get a lot of responses so we might not be able to include all the submitted images in the final SIGGRAPH paper.

We will try to include the best ones in our final paper, so if you have something interesting and would like to see it at SIGGRAPH, please email it to me at psen AT ece.unm.edu

You can send us several images if you have them, we will choose the ones that look the most compelling. If the file size is large (> 5MB) which is most likely the case, you can put your images somewhere on the web and send us the link to them.

Thank you very much for your help.

Sincerely,

Pradeep Sen
UNM Advanced Graphics Lab