Gopro Footage resize & stabilize - best workflow

Question

Hi maybe some of you AE pros can help
I'm wondering which is the best workflow for stabilizing 2,7k gopro footage while resizing at the same time to 1080p.
This is not clear to me as Im not familiar with how AE handles resizing internally.
When I put the original footage into an 2,7k composition
stabilize it there and put the whole composition into another one at 1080p - will this be a loss in quality?
Is it better to start off with a 1080p comp?
greetz, David

Answer 1

Any time you resample information you are loosing information. Scaling down is visually more pleasing than scaling up, but if you want to optimize the look of the video you should add some kind of sharpening. You cannot judge sharpening, like unsharp mask, until you render a full resolution preview and look at it at 100% because too much detail in a video is worse than too little. You'll get all kinds of unpleasant ringing in high detail areas. The same thing goes for upscaling images. To get the best results you have to apply the right kind of scaling algorithm and do some filtering to hide the upscaling artifacts and make the image more pleasing to look at. There is a limit on how far you can scale up and still get a pleasing image, there is also a limit on how far you can scale down and still see detail. For example, if you had 4K footage of a house made of bricks and you scaled down the image to 100 pixels wide it is very unlikely that you would still be able to see individual bricks.
Here's the deal with manipulating pixels. Everything is built on a grid. If you move your original pixels in any way so that they do not line up perfectly with the grid the pixel values are then interpreted and detail is lost. Draw a vertical one pixel wide white line on a black background then move the line 1/2 pixel to the left or right and you don't have a white line any more, you have a 2 pixel wide gray line. Just the act of stabilizing your footage causes a slight quality loss because it is moved in sub pixel increments and the pixels have to be re-interpreted.
If your final product is standard HD, then that's all the quality you can get. There are fewer pixels so there is less information and less resolution, but an HD sized comp will easily fill a 60' wide screen in your local cinema and be extremely pleasing to look at as long as you are not standing 6 feet from the screen. You're obsessing too much about this whole process. Build your comp, do your work, then check to see how it looks. Make a single frame look as good as it can and render a test to see if it is good enough when it moves. That's the only sure fire test of quality. Then compress for delivery using the highest data rate allowed with multi pass rendering and see if that quality is also good enough to pass your quality control standards. At the end of the day you must realize that nobody watching a movie is going to analyze every pixel on every frame. You would be surprised at what some still frames from big budget Hollywood blockbusters look like. When the image is moving, you can get away with things that would never pass in a still image.