|
|
|
|
|
|
by spoiler
3613 days ago
|
|
|
This does look fascinating. The video is a bit showbusiness-like (for a lack of better expression). A few questions if the author is on HN:
I wonder how long it takes to analyse a 5 second video? Also, it seems the algorithm only works with static images after the initial video analysis, or am I wrong? Also, how long does it take to render the "virtual" state of the object? |
|
|
The simulation runs in realtime, and is interactive. Generating the simulation can take a while though.
The 5 second video took about a minute to process on my laptop. But longer videos, and higher-resolution videos, can take much longer. For instance, we've used the technique on high speed video to recover audible vibration modes, and this can take hours to process because the video itself is so big.
I wrote the processing code in Matlab, and the simulation in c++ and GLSL. The simulation is pretty well engineered because I wanted it to run in realtime. I didn't put a lot of time or effort into optimizing the video analysis part though, so it probably could be made faster.