Point Tracking Gets Smarter in Silhouette

Added: 2026-06-03

[00:00:00]When it comes to tracking in Silhouette, you have the tried and tested point tracker, planar tracking, Mocha tracking, optical flow tracking, object ML tracking and of course, 3D tracking.

[00:00:12]These are all amazing technologies and a mandatory component for roto, painting, and compositing.

[00:00:19]As part of our ongoing effort to make things even easier for you, we decided to revisit the classic point tracker and give it some ML love, which I'll think you'll appreciate.

[00:00:30]I'm Grant Kay for Boris FX and let's check this out.

[00:00:36]The point tracker has been a fundamental tracking tool for decades and is a staple for any roto, painting or compositing workflows.

[00:00:45]Typically, you'd place a point on your image that you'd like to track and as long as the pattern or reference is consistent and visible, you should get a decent track.

[00:00:55]Realistically, reality never works that way all the time and you might be tracking an object going off screen or something gets in the way of the reference.

[00:01:05]You can overcome this with manual intervention but we thought ML might be able to help in these cases.

[00:01:12]Let's take a look at a couple of examples.

[00:01:15]Here we have a skateboarder and he's weaving on the path with his bottom half going on and off screen.

[00:01:22]Now as part of a paint or roto task, we need to track this part of his shoe.

[00:01:27]Let's look at how the classic point tracker handles this.

[00:01:31]Using OPTION on Mac or ALT on Windows, we'll click and add the point tracker.

[00:01:37]We'll also make the boxes slightly bigger to get a decent reference pattern.

[00:01:42]Ensure the mode is set to Classic and press the TRACK button.

[00:01:48]The analysis begins and we can see the keyframes being created.

[00:01:54]Now before we even get to the edges of the frame, the reference is changing and loses the track.

[00:02:00]No point in letting this finish, we'll just stop the analysis.

[00:02:04]This is a typical weakness of the point tracker.

[00:02:09]Let's undo this and try the Point Track ML.

[00:02:13]The feel on the analysis will be slightly different but the point tracker is locked on.

[00:02:19]This is because there is now a large awareness grid around the point tracker and the ML is trying to figure out what it is tracking in order to deliver a locked track.

[00:02:30]Even with distractions like going off screen, occlusions and motion blur, this is not a 100% guaranteed solution but the point tracker now has a much higher chance of success.

[00:02:43]To validate this, we'll turn on the viewer stabilization and lock it to the point tracker.

[00:02:50]Looking at this result, we definitely have a much better track that can be fed into the rest of the compositing pipeline.

[00:02:58]Now going off screen is one situation.

[00:03:02]Another common challenge that happens all the time are things getting in the way when tracking.

[00:03:08]I'll switch over to another session and play the next example.

[00:03:12]This time we need to track the buckle on the lady's bag.

[00:03:17]Zooming into the shot and taking a look at the frames, we have that common situation of hands and straps getting in the way.

[00:03:25]Let's first use the classic point tracker again to see what happens.

[00:03:30]We'll add the tracker to the shot and resize it for the bag and buckle.

[00:03:36]Once again, remember to set the mode to classic and start the track.

[00:03:41]As soon as the hands get in the way, the track is lost.

[00:03:45]We'll undo the failed track and switch the mode to Point Track ML.

[00:03:51]Now before we track the shot, I would like to mention the accuracy pull down menu.

[00:03:56]The accuracy settings use different options with the AI to understand the track.

[00:04:02]By all means, try the different accuracy settings but "High" normally gives the best results and is set as the default.

[00:04:11]We'll kick off the track and see what happens.

[00:04:15]As the hand and strap cover the bag and the buckle, the Point Track ML holds onto the track and produces a very decent and usable result.

[00:04:36]We'll turn on viewer stabilization and lock it to the tracker.

[00:04:41]That's pretty solid and it saved us a lot of time.

[00:04:46]Silhouette brings in the point tracker of the future, adding one more tool for this invaluable and inescapable VFX task.

[00:04:54]If you want more, let us know in the comments.

[00:04:57]We also have the official Boris FX forums and Discord server for bigger discussions.

[00:05:02]Links below.

[00:05:04]I'm Grant Kay for Boris FX, and see you next time.