| Author | Message | ||
     Rrrob (Rrrrob) Senior Member Username: Rrrrob Post Number: 230 Registered: 5-2006 Rating: N/A |
Anyone have experience with this? Curious how they pull it off.... | ||
| Scott Warren (Scott_warren) Member Username: Scott_warren Post Number: 43 Registered: 8-2006 Rating: N/A |
Here is a good example of the overview:http://www.ee.cityu.edu.hk/~lmpo/publications/2010_3DV_Conversion_Seminar.pdf | ||
| Rrrob (Rrrrob) Senior Member Username: Rrrrob Post Number: 231 Registered: 5-2006 Rating: N/A |
interesting read, but it doesn't cover what method DVDfab is using. I have seen a lot of 'conversions' where, for instance, portions of the screen are displaced to create a parallax difference, but the effect's success is hit or miss (usually miss), as that method assumes foreground objects are always in the bottom 1/3 of the screen, background images are in the center, etc. | ||
| Larry Elie (Ldeliecomcastnet) Senior Member Username: Ldeliecomcastnet Post Number: 150 Registered: 10-2006 Rating: N/A |
It's all about the depth map. If you look at the color channels of a lot of still, you find trends. Wow; I can use that for a depth map. Well, maybee. The problem is that when you or I look at the color channel, we can tell what belongs in the front or back, and agree or disagree with it. It worked because of lighting, or object color. That's hard to completely automate. I think most of the conversion algorythms for still today look at saturation, at least as one parameter. Saturated objects appear to be foreground. Before you say that's silly, look at a saturation map; it looks like a depth map many times because the object of the photo is often in the foreground. Again, it doesn't always work. But we are talking movies here, and that is actually much simpler. Much. Why? because of all the compression already used. Yep. For most video conversions, relitive motion is used. Objects (groups of pixels that didn't change from one frame to the next) that are moving rapidly are assumed to be foreground. Objects that don't appear to be moving are assumed to be background. The real trick is to determine what is an object, and the people who did the compression are assumed to have done it right. Will that always work? Of course not! An object moving toward you is an example, or a fast moving object in the background, so they put a delay buffer in, which helps. But if you look at the algorythms of Sony (who's first blu-ray converter has been demonstrated) or Phillips, that's the heart of the conversion. That said, the problems are all in the details. Is that what DVDFab did? I don't know. But I would be shocked if it was much different than that. | ||
| Rrrob (Rrrrob) Senior Member Username: Rrrrob Post Number: 232 Registered: 5-2006 Rating: N/A |
i just don't see an 'automated' process working, ever. I've done still depthmaps that look pretty good (http://www.flickr.com/photos/23816910@N04/sets/72157624062105183/), but video is indeed a completely different animal. (Message edited by rrrrob on May 21, 2011) |
