Recently there was a nice article on FCP.co regarding testing stabilization plugins for Apple's Final Cut Pro X. The videos from that test were published to YouTube so I thought it would be interesting to see how MotionBend's output compared to the other products - using content thought to be objectively suitable for testing.
I downloaded the videos from YouTube via a browser plugin. This resulted in four videos with test material. Each video was downloaded as 720P25 and includes the original footage and the comparison videos from each product. There are also decompression-compression artefacts from using content obtained from YouTube so strictly speaking, this isn't a fair test.
You can find the four videos used for testing here:
1. The video was opened in MotionBend. You can download MotionBend and use it in evaluation mode if you haven't purchased it yet.
2. The video was scanned to find the end of the original footage, located at approximately the 5 second mark.
3. The end of the selected frames range was set to the playhead position by hitting ⌘+] (Command + Right Square Bracket). Alternatively you can drag the small red triangle or use the context menu.
4. The problem with using these videos is that they have captions that will confuse the analysis process. Before we analyze and correct for stabilization we're going to mark foreground regions that should be ignored. Start by selecting "Regions" mode, under the frame selector.
5. Draw a box around the "Original Footage" caption by clicking on the preview and dragging a rectangle around the caption.
6. The regions mode was changed to "Edit" in the "Foreground Regions" section on the right hand side.
7. Click on the box to select it. Drag it to adjust the position if necessary or click on the corners to adjust the size and shape.
8. Now press "A" to make the region apply to all frames. The box changes color to cyan.
9. Click on the Analyze button.
10. After a short wait, the analysis is finished. Simply hit the play button to view results.
Unfortunately the caption now moves with the motion of the camera that we were trying to remove. We think the MotionBend result compares very well against the other methods but we'll let you be the judge of that.
If you do this comparison you may get different results depending on what resolution video you downloaded and what quality YouTube served to you. If compression artefacts are visibile then they will affect the motion analysis.
MotionBend allows you to change the look of the camerawork by editing "motion events". In the screenshot below we deleted an event, adjusted the start velocity of the second horizontal event and modified the amount of acceleration and deceleration.
This video is a little challenging because most of the background is out of focus. The motion analysis is being confused by the sharply focused Meerkat in the foreground and the out of focus background, this is the main cause of the "pulsing background" referred to in the FCP.co article.
To generate a stabilized video where the Meerkat's movement is visible, all the foreground was marked. This included the caption, the Meerkat, its paw and another moving object to the left of the caption (maybe a tail ?). We used four regions in total, as shown below. After marking them and setting them to apply to all frames as in the workflow above, we played the video and adjusted the regions to make sure all of the Meerkat was covered before clicking Analyze.
In the resulting video, you can see the Meerkat's movement. To further emphasize its motion the global motion can be set to "Motion Cancelling" for full lockdown.