
This is an article about knowing, rather than hoping, we’ve nailed a track with Foundry’s Nuke. The approaches we’ll discuss are not software-specific, but our examples definitely will be. Feel free to apply these ideas to your compositing software of choice.
A Case Study

I had an experience during dailies for The Great Gatsby that changed my approach to tracking forever. The shot of that beautiful, yellow car careening around a corner came up for review. The artist on it – I remember it being Ben McEwan, but could be mistaken – had replaced the heads on the stunt driver and passenger with Gatsby and Nick. With the big, dynamic car motion, it was hard to tell if the faces were really stuck on. I would have relied on my comp supe to say if it was good enough. But Ben had added two little squares in the bottom corner of the frame containing stabilized views of each person’s head. Without the overall motion, it was really easy to see if his patches were truly stuck on to the plate.
My approach at the time relied on someone else to find my faults. Ben provided tools for verifying that he’d gotten it right. It’s the difference between hoping we’ve nailed it and knowing we have. I learned my lesson (thanks Ben!) and haven’t assumed a track was stuck on since.
Checking Tracks
Maybe the Tracker is showing green lights and the patch looks stuck on as the feature we’re tracking thrashes all over the screen. Watching the overlay as it follows our tracked feature is better than nothing, but still suffers from a weakness in human perception: we don’t distinguish small slips on things that are making big movements in our visual field.
So how can we know our track is really stuck on? We stop it from making big movements in our visual field. Inverting our track yields a stabilize. Apply that to the plate.

Now… is that feature really nailed down… or is it wobbling a little bit?
This is the time for a gut-check. We’re about to build a lot of work on top of that track. It’s much easier to be sure we nailed it now, before we use it to drive a bunch of things. The last thing we want is to have our supe call out that it’s slipping when we’re up against the delivery deadline and have to remember to fix it in every place we use the track. So let’s be sure it’s locked down now.
To increase the power of our slip-detection, lay a reference grid over the stabilized footage. Then crop it down to only the feature we’re tracking so it caches quickly in Nuke’s Viewer. Use the expression-linked stabilize so our tweaks to the track appear instantly.

Fixing Failed Tracks
If our track fails the stabilize test, don’t stress, but also don’t quit. We have options. Try one (or all) of the following.
Try Again with a Different Approach on the Same Feature
Nuke’s Tracker has a lot of options for how to find a feature over multiple frames. Try re-running the track with different settings. A common one is to change the warp type to better match the kind of transformation that’s happening to our feature in the plate. (Affine is a good option for features that are changing their perspective.) We could also try setting initial keyframes, have the Tracker adjust for luminance changes, use a higher track tolerance, or grab a reference frame more frequently.
Break It into Two (Or More) Tracks
For shots that have a definite change in light or motion which is causing track trouble across the transition, we could try using a separate tracker for each part.

Post-track Tweak
If the track is stable, but doesn’t lock down the feature we need it to, we can add a manual tweak downstream to get it running where we need it. This is often a matter of putting a Transform downstream and setting keyframes at the beginning and end of the shot (and maybe one somewhere in the middle if necessary.)

Track Something Else
Maybe we’re having trouble tracking the feature we want. If there’s a feature that’s easier to track nearby, something that moves like our preferred feature, we can track it and use the same post-track tweak approach to adapt it to what we need.
Manual Tracking
Avoid manual tracking if at all possible. It is slow and painstaking, but sometimes only a person can make sense of the chaotic motion of a feature. We’ll do this only as a last resort, after every other approach has failed. A careful review through the stabilize after each pass through the track shows us what to tweak on the next pass.
Using the Track
Now that we have our track nailed down nice and tight, there’s something we need to discuss before we use that track data. I know I’ve been harping on making sure the track is perfect before we use it. But this is the part where we consider that maybe we’ve made a mistake in the track (perish the thought) and look at ways to protect ourselves from further damage.
There are two ways to use the track we just made. We can expression link the track data, or we can bake it out as keyframed curves.

The expression link has the advantage of updating as soon as we make a change to the track. This is convenient if we do find an error in the track and need to fix it. But it’s really not good if we’ve based some roto or paint on the track data. Updating the track moves our splines or strokes from the places we put them, introducing errors where previously there were none.
Some folks prefer using baked keyframes just because they don’t like to have green lines stretched across their node graph. That’s fair, too. (Although we can easily hide them by typing alt-e.)
Consider these possibilities when using the track. Expression links are generally safe for simple things like backgrounds or static patches, but baked keyframes are much safer when we’ll be building a lot of work on top of the track.
Supporting Tools
Earlier we discussed track solutions that will end up as multiple nodes. This can work perfectly fine as long as the nodes stay together. Remember that transform node concatenation in Nuke only works if all the transforms use the same filter mode. Motion blur will be determined by the last node in the line.
For folk who really hate the way this looks or have a situation where that can only accept a single set of transform values (such as iTransform), have a look at Nuke Matrix Toolset by Erwan Leroy and Mathieu Goulet-Aubin. It contains a tool for baking multiple transforms down into a single node. It’s very user friendly. It also provides an easy way to make Roto nodes that are set up for track-assisted roto. There’s so much more in the toolset, but these are the things I reach for every single day.
Go Forth and Nail Tracks
Now that we’ve learned how to check tracks and several ways to fix them when they’ve gone wrong, we’ve immunized ourselves from bad, track-related surprises during dailies and tech-checks. We’ve also started down the road to being a trusted member of the comp crew. (Rule 3) Don’t be surprised when your supe notices your increased reliability and starts handing you harder shots. Don’t worry, these techniques work on hard shots, too!