Clearing the cache remove tracking data

Clearing the cache remove tracking data of the magic mask (beta 2)

[Rohit Gupta]

This is expected as of now. When you track, it actually stores the mask in the cache folder.

We'll see if we can improve it.

Please note that if you change your timeline resolution, you will also need to re-track the object mask.

This is expected as of now. When you track, it actually stores the mask in the cache folder.

We'll see if we can improve it.

Please note that if you change your timeline resolution, you will also need to re-track the object mask.

Same thing with when anything is changed in the Inspector?
It does that when making the clip a compound clip.

But making a compound clip first, then going into it, do the tracking, go back on the main timeline, then it works

I found the magic mask data to be very fragile. In certain cases an undo done several steps after the track erased the tracking data, and a redo didn't bring it back.

This is expected as of now. When you track, it actually stores the mask in the cache folder. We'll see if we can improve it.

Yes please! I clear cache frequently and definitely do not want to lose any actual work like tracking data.

This is expected as of now. When you track, it actually stores the mask in the cache folder. We'll see if we can improve it.

Yes please! I clear cache frequently and definitely do not want to lose any actual work like tracking data.

I had to track again so many times Because I moved something, change the cache style (or deleted it).

I watched videos on youtube about the magic mask, many people are not on specs out computers. When I saw 6-7 mins to track a 10 secs video, I can't imagine the frustration if the track data is lost.

I had to track again so many times Because I moved something, change the cache style (or deleted it).

Yeah, I saw a video saying that moving the clip will delete the magic mask tracking data and they suggesting putting it in a compound clip because the compound clip can be moved without affecting the tracking data... but that doesn't prevent deletion when clearing cache. Cache should be temp files that are safe to delete because they will be rebuilt when needed. Tracking data certainly does not all under this category and should not be stored with cache files.

I delete the entire cache folder to clear cache for all projects at once (much faster the opening each project to clear it). Please don't save any important data inside the cache folder.

This is expected as of now. When you track, it actually stores the mask in the cache folder.

If the mask data is missing, it must re-track automatically during playback if possible (or at least display a warning message and offer an easy way to retrack all masks in the project automatically).
Another solution is to add a project setting so the user can choose where the object mask data is stored.

Otherwise this gonna be a big problem, especially if you loading some old projects and you realize that all object masks are missing and you have to search and re-track all clips one by one.

Another solution is to add a project setting so the user can choose where the object mask data is stored.

I'm not sure what is different about the object mask, but I wonder if they could simply save this tracking data inside of the project file. That seems to be the way other tracking works, I've shared projects using tracking in Fusion and all of that was included in the single project file. All the tracking was functional when opening the project. If you have to specify a different location for object tracking, then you'd also have to send all of these separate tracking files when collaborating with others.

Yeah no track data is lost for power windows, or the other tracking tools. In the color page or Fusion

The magic mask tracking data includes the mask so it's quite large in comparison to tracking data for power windows and effects. That's probably why they store it outside the project but I agree that it should not be stored in the cache folder (that was a really peculiar design choice). Maybe in the capture folder? I don't have an answer but wherever it is stored should follow project archive media, etc. so that it is not so easy to lose the mask data.

The magic mask tracking data includes the mask so it's quite large in comparison to tracking data for power windows and effects. That's probably why they store it outside the project but I agree that it should not be stored in the cache folder (that was a really peculiar design choice). Maybe in the capture folder? I don't have an answer but wherever it is stored should follow project archive media, etc. so that it is not so easy to lose the mask data.

What I generally do now, I create a compound clip from the start, then I open it and do the magic mask stuff there. When doing that, the compound clip nesting the tracked clip can be moved, resized, etc on the timeline.

If we track, then make a compound clip out of it, the track data is lost.

When a project is closed and re-opened, if nothing moved, the track data is there.

I don't think it's a size problem. I'm sure it's still under development, but with the response from BMD, "as of now", it can be both. The behavior is what's expected, or it's just not finalized

I imagine the mask would be quite large (it's media, after all) so the size is a reason not to store the mask within the project.

I imagine the mask would be quite large (it's media, after all) so the size is a reason not to store the mask within the project.

The depth map tool doesn't lose created mask

I imagine the mask would be quite large (it's media, after all) so the size is a reason not to store the mask within the project.

The depth map tool doesn't lose created mask

Are you saying that the depth map tool creates a series of masks saved to the project (where a series of masks is an N fps video file of greyscale images)? Could it be that the depth map tool dynamically creates the mask (and doesn't cache it as a motion graphics file)?

If it does generate a motion graphics file at the FPS and resolution of the timeline, then where is it stored?

How the tools work behind the scenes don't matter because both of us don't know how the software was coded.

The fact is, track data is lost.

1) When a clip is transformed into a compound clip, you open it... then use the magic mask on the original clip, no track data is lost when this compound clip is move, resized, etc. on the main timeline.

2) Track data is lost when the magic mask is directly used on the original clip, and this clip is moved, resized, transformed into a compound clip, etc.

But I said that before in this thread.

So there is no problem of "mask being too large" to be stored or whatever.

Likely we are speaking of cross-purposes.

Even if you use a compound clip, you will lose the mask data if you purge your cache, presumably, but I get your point - if mask data survives in a compound clip during all the other manipulations, then it should be possible to have it survive on the native (non-compound) clip as well. Hopefully we will see improvements to how the magic mask is implemented over the course of the beta cycle.

Big PLUS and BIG 1. Magic Mask is great, but sooo fragile. Making me nuts. I'm using the depth mask whenever I can, in place of magic mask -- when I would much prefer magic mask if only I could trust it to hold itself together -- just so I can be confident that I won't come back to a clip and discover that the track is gone. Again. Or worse, won't discover only after an export that a "magic-demasked" node borked a clip's grade. Working downstream often "demasks" upstream for some reason.

Last night I had a dream about magic mask -- a nice one -- that come morning erased the track data on a real-life node. Today I breathed on a node and its magic mask tracking disappeared. My dog took a nap in the sun and erased my tracking again. Fragile.

So, plus 1 to the "fragile" magic mask track data many are talking about in this thread! Cheers.

Making me nuts.

Welcome to the club ahah.
I'm sure this tool will be so good in the future.
I use it so much now (even with its current flaws).

I also had tracking data lost on seemingly innocent changes. If this was "as designed" and not a bug, I think Resolve should ask user before erasing tracking data, as it can take minutes or hours to re-track some of them. (With of course, being able to turn this confirmation dialog off in the prompt and in settings.)

So, I had a node prior the one on which I wanted to use the magic mask.
On the first node, I added contrast and upped the gain a lot so the magic mask can really find the edges.

But the track data is lost if you reset the prior node.

And adding a stroke when it's on "better" doesn't work

Just like tracking data for power windows (or Fusion projects), the data should be set in stone until the user changes it, even if the media source changes (including resolution!). One expects node trees and nodes to remain stable -- and in Davinci Resolve they DO stay stable, until the introduction of the new nifty Magic Mask feature. Unfortunately it is not tenable to send outputs to QC, only to discover that a clip grade is bad solely because one magic mask did not "stick".

Had a thought. I'm presuming part of the problem is that instead of tracking data, this tool produces alpha video instead. Is that correct? If that's the case, I'm assuming then it would be unwieldy to save that video in the project file. But the lack of durability of the mask is a problem.

I'd propose that instead any mask alpha video be saved with project media, in exactly the same way render-in-place results are stored with source media rather than cached (thrown into cache directories). For me personally as a DR user, this would be rather huge. I wager it would be a big improvement for others too. .02, if I'm on track with the problem here.

Thank!

Until the cache problem is not solved I use the following method:
After changing the scene I use Render in Place and store the result in a folder in my project folder.
The "Render in Place" scene you can always bring back by using "Decompose to Original"

Dirk PEL

With 18.0.1 I'm not seeing clearing render cache affecting the magic mask. I think they actually changed where they store the mask data since the point at which they informed us of the unfortunate design decision. That's not to say that the magic mask can't be invalidated quite easily by changing any aspect of the source (including changes on the edit page or in an upstream node on the color page).