THIS IS ADDITIONAL INFO TO THE ARTICLE AND FILE:

My first encounter with Softdistortion was from a course with Victor Wolansky at Fxphd.

I had previously attempted to work on the shot, albeit a different ‘take' from the same shoot.

When I initially saw this shot, my first reaction was, this will be easy, it's long but there is plenty to track.

I could see that motion blur and lens distortion was minimal, and only a little rolling shutter needed to be corrected, which The Foundry's Rolling Shutter plug can deal with easily.

Upon working with the shot it was immediately obvious that it was going to be a very tough shot to solve, primarily due to its unusual and colossal near 3000 frame length!

I knew that any initial minor mistakes, would gradually throw the entire solve off by the end, so it was going to be imperative that it had good trackers throughout.

My first task was to remove the rolling shutter distortions, then to tackle the lens distortion. Although I was supplied with a typical front-on lens distortion grid, it wasn't quite shot correctly as the camera wasn't close enough to the grid. This meant that the grid didn't fill the entire image, and didn't show the distortion at the edges of the screen, which is where it's usually most prominent. Knowing this, I did the best I could work out but it was a little tricky as plenty of the surfaces aren't quite as straight as they first appear.

I started working with the shot in PfTrack but soon realised that simply due to the length of the shot, it was going to be a long and convoluted process, as the program is only 32bit so it cannot cache all of the frames. Since PfTrack works frame-by-frame, doing some initial auto-tracking proved to be a slow process.

I asked a friend if I could use his workstation which had Syntheyes installed on it, and I thought I'd see how I'd get on with the Softdistortion shot using that.

Much to my surprise Syntheyes, even in its 32bit form, managed to run an auto-track pass without a problem. With an extremely long shot like this it's a good idea to use both auto and manual trackers to help calculate the solution, primarily to save time but other issues arise when just using manual or auto-tracks.

Syntheyes uses completely different algorithms to other tracking software, and it doesn't need to run frame-by-frame in auto-tracking. It merely took a couple of minutes, whereas PfTrack had taken over an hour as I could only load around 1300 frames at a time into the cache.

After cleaning up the auto-tracks, it was then time to create the manual-tracks. I ended up with around 15 manual tracks throughout the shot, it would have been easy to have 80+ but I decided to just have a few to help refine rather than force the solution.

Upon reviewing the shot, I could see that it wasn't 100% perfect but acceptable to work with. This is largely down to not quite getting the lens distortion fully un-warped, and so this caused some of the trackers to not be in exactly the correct place.

I am used to working with PfTrack, as 99% of the time it can handle most shots but in this particular case, the unique way in which Syntheyes tracks and calculates, it was definitely the better solution. At the time PfMatchit had only just been released, thus I wasn't as comfortable then as I am now with it (an utterly fantastic program). PfMatchit would have addressed the RAM limitation that a 32bit program such as PfTrack has, so I'm quite keen to go through the shot again to see how it copes.

SD wrote: \"For this shoot the lens was Canon EF24-70/2.8 L USM Lens. Unfortunately there is no dist chart or record of focal lengths used." :(

Here is an initial test : VIDCLIP

Obviously its not refined yet but a good indicator that it should solve ok.