Hey Nicolas, Thanks for the tips. I was able to get a better result from the video I made a while back. I realized I did not use the "Previous Frame Same Block" option originally. This made a huge difference. As I didn't have much time to fuss with it, I lowered the translation search steps to 128 as you suggest.
Here are my results with comparison of before, after and zoom. The video is very compressed, but you can see that the zoom doesn't degrade too badly, as my source is HDV, which is something you'd be interested in seeing: http://content.serveftp.net/video/motionexercise.m2v 7MB scott On Wed, 2006-10-18 at 18:27 +0000, [EMAIL PROTECTED] wrote: > Nicolas, > Great analysis. This will help me. I will try to get to this > tonight/tomorrow. > > Do you have before and after vids? > scott > -------------- Original message ---------------------- > From: Nicolas <[EMAIL PROTECTED]> > > Hello, > > > > I did some tests with the motion effect, and I got a result which > > satisfies me. In fact, that effect was easier to use than I thought. > > Here is some feedback about it. > > > > I used the motion effect to stabilize DV footage. The camcorder was > > installed on my helmet while I was following a group of motorbikes. My > > motorbike's shocks are in good condition and the position of the > > camcorder helps to reduce vibrations. However, the road was quite bumpy > > and the footage is finally still jerky. My camcorder includes an optical > > stabilization system (Panasonic OIS), but it's not that efficient in > > such harsh conditions. > > > > Firstly I selected on the timeline the part of the footage I wanted to > > stabilize, using the in and out points. Then, I applied the motion > > effect on that part of the video. I did not use 2 tracks (top and > > bottom). Motion tracking was done and applied on the same track. > > > > I tried the default options. I used the following parameters to define a > > block on one of the motorbikes which was in front of me: > > - translation block size > > - block X, Y > > And I rendered the video. Even if my computer got more than 10.500 > > bogomips (dual-core over-clocked AMD64 X2 4600+), it was not convenient > > to preview the result in the compositor. That is normal and described in > > the official manual. Since the option "Draw vecors" was enabled, I was > > able to see the block position and the vector shape on the video. > > > > I noticed the block was following the motorbike, and the picture was > > stabilized. That was a good result for a first try. However there were > > some problems. At the beginning of the video, the centre of the block > > was positioned on the plate of the motorbike. Sometimes that centre > > quickly moved to another part of the motorbike (the rear passenger > > handle, located just behind the pilot). That resulted in the frame being > > suddenly translated by a vector of several tens of pixels. Therefore the > > video was quite stable, but sometimes there were large and quick shakes. > > > > I tried again and changed some parameters. I decided to select the > > "Previous frame same block" option. That option is recommended for > > stabilizing jerky camcorder footage. Its goal is not to "follow" an > > object. The block stays exactly at the same place during all the effect > > length. I enlarged the block and selected almost all of the upper half > > part of the video. I also selected the "Stabilize subpixel" option. That > > gives a finer stabilization. Finally I also rduced the "Maximum absolute > > offset" value to limit the stabilization amplitude. I prefer to get a > > non-perfect stabilization on some places on the video, than having a > > very large black border on one side of the picture during big shakes. > > > > I rendered the video and got a better result. The image was stable and > > there were no sudden translation anymore. Since the result was good, I > > re-rendered after having deselected the "Draw vectors" option. The block > > and vectors were not drawn anymore on the video. > > > > I then imported the resulting rendered .dv file into my project. I had > > to zoom in the picture, in order to get ride of the black borders > > generated around the picture. I had to tweak the projector keyframes. > > > > During my tests, I noticed increasing the "Translation search steps" > > increases a LOT the rendering time. You can set it from 128 up to > > 131072, and the rendering time is a lot longer when using the maximum > > value. I reduced that value down to 128, and I compared the result with > > what I got when using the maximum value (131072 steps). It is indeed > > very easy to compare the result, since the translation x and y values > > found by the motion effect are displayed in the console. What I did was > > copy/paste the console output to text files, and compare them using the > > diff utility. There was absolutely no difference at all in the result. I > > then set the "Translation search steps" value at 128. I assume > > increasing the value can be useful in some cases, but I do not know > > which ones. > > > > The main problem I had was working with the stabilized video. The black > > borders force one to zoom in and define projector keyframes to move the > > projector around the screen. Since my original footage was quite shaky, > > the black borders were large at some places. Moving the projector > > implies having some camcorder movement, but since the projector > > movements are slow, the result is far better than the jerky original > > footage. The more your footage is jerky, the more you have to zoom in to > > discard the black borders. That is why the result should be better with > > high-def footage than with "normal DV" footage. I also had some problems > > to stabilize panning movements. A large black border often appears on > > one side of the video. > > > > Nicolas. > > > > _______________________________________________ > > Cinelerra mailing list > > [email protected] > > https://init.linpro.no/mailman/skolelinux.no/listinfo/cinelerra > > > > _______________________________________________ > Cinelerra mailing list > [email protected] > https://init.linpro.no/mailman/skolelinux.no/listinfo/cinelerra _______________________________________________ Cinelerra mailing list [email protected] https://init.linpro.no/mailman/skolelinux.no/listinfo/cinelerra
