Palacono's Bugs'n'Misunderstandings [#52 Hitfilm Framerate issues (Link only)]

1468910

Comments

  • Palacono
    Palacono Posts: 3,423 Enthusiast

    New Bug/feature: #24 Set Matte Weirdness
    -------------------------------------

    Short and sweet. Set Matte still has a (minor) bug I reported last year with noisy edges and it also goes freaky in 3D, which it also did last year, but I didn't get around to reporting.



  • Palacono
    Palacono Posts: 3,423 Enthusiast
    edited October 2016

    New Bug/feature: #25 Dependency Bug?
    -------------------------------------

    Parts of this has shown up before; which is when making a change in an embedded composite shot: when you go back to the shot it is embedded in, the changes don't show up until you scrub the timeline a bit to make it refresh the viewer image...even although it's already refreshed the image, because it's had to change it from what was just displaying in the embedded composite when you were making those changes! 

    It uses out-of-date information about the frame without doing the calculations it should do when you've changed something that one of the layers contains. i.e. an embedded comp shot.

    The 'dirty info' list isn't being updated, basically.

    That's annoying, but not too destructive; you quickly get used to pressing . then to go forward and back a frame to get it to wake up. ;)

    The other thing is more of a problem. Two layers that should be completely separate get joined at the hip and produce wrong results.

    I did try another variation where I duplicated the layer in the Media Bin and dragged that into the comp and then they do remain separate; but it shouldn't be necessary and I think this is just common or garden buggy wuggy. :)



  • Palacono
    Palacono Posts: 3,423 Enthusiast

    New Bug/feature: #26 Slow Delete
    -------------------------------------

    This has frustrated me forever, but it's only occasionally that I delete more than half a dozen keyframes, so that fact that it's so slow doing that is something I generally live with. But with HF5 about to drop real soon now and the latest HF4 update being so...eh? , I'm hoping it's not too late to add this to the pile for HF5.

    Deleting keyframes is slow. Really slow. So slow it's like every single byte is being deleted one at a time with full callback checking to the slowest Windows library routine ever written that's sitting on top of an old Int 21h DOS call. 

    If it's calling a library routine for every byte read/written, that needs a cache in there to speed things up. Back in the DOS days, you could get about 100000% speed increase by adding something as small as a 32 byte cache inserted ahead of all read and write routines because Microsoft's code is bulletproof but sloooooow.  A Read Cache will auto fill on first use and if you don't want to have to remember to flush the Write Cache when you're done you make something clean up after you're done automatically.

    Or, perhaps Hitfilm is doing something weird(?) like reprocessing the entire Composite shot (and any embedded ones) after every single keyframe is deleted.  Even then: there was not a lot happening in that comp that could slow it down that much. It plays faster than the time it took to delete it.

    This was deleting about 20 keyframes of about 20 points each, so 400 data items, which isn't a lot at all. But if the time taken to process them is so slow, when you start to add more layers with keyframes on them: that overhead starts to mount up throughout the project. 

    Could just general data handling be one of the reasons why Hitfilm is slow and which no amount of GPU/CPU/RAM updates will help?

    Same applies to tracking keyframes and they don't even have any dependencies to speak of that would need adjusting (which other, parented points with Keyframes could have), they're just a point on the screen...

  • NormanPCN
    NormanPCN Posts: 4,082 Enthusiast

    I enjoy watching your thread/findings. This one can be fun to speculate on so...

    I have doubts it is memory management related. I have seen developers do strange things with memory management over the decades but they pretty much use the compiler RTL memory routines or the OS routines. Neither one of those are brain dead.

    This is something I wonder about. When you delete a keyframe the values of intermediate frames change. So maybe Hitfilm runs the keyframe timeline recomputing these values at frame positions. No graphics. Just the keyframe value at each frame. Now suppose you have a subroutine which deletes a keyframe and which runs the keyframe timeline recomputation. Kinda useful to create a reusable subroutine. Okay, now what if when you delete many keyframes. This routine simply runs in a loop calling the delete subroutine once for each keyframe to delete. We now have an n-squared computation. We needlessly re-compute the timeline n-1 too many times. The proper thing in this circumstance would be to re-compute only once after deleting all selected keyframes to be deleted.

  • CedricBonnier
    CedricBonnier Posts: 1,197 Staff

    @Palacono Thanks for reporting this, it is an issue in the Viewer (go to the editor sequence and press undo / redo and the operation is much faster). We will look into it but this has always been behaving the same (I quickly checked in HF2 but I guess it's no different in HF1).

  • Palacono
    Palacono Posts: 3,423 Enthusiast
    edited October 2016

    @NormanPCN thanks, I could do these all day every day if I had the time, but it's only when something a) ticks me off more than half a dozen times and b) I have some spare time and think I can reproduce it fairly quickly that I bother. As it is, each one takes about an hour to put together, upload and write about, so they get semi-randomly cherry picked.

    I like your n^2 theory, or reusing something because it works well and only later finding the overhead's a bit more than anticipated when you multiply it up.

    But as @CedricBonnier is at least aware of it, maybe he'll find other little v1.0 "Just ship the program!" code snippets lurking, and HF5 will come out of the traps like a whippet strapped to a rocket! :D

    I still feel like there is a:
    "for n=0 to number_of_possible_effects_and_lighting_models_that_can__be_applied; check_pixel;
    next n;
    "
    loop in there somewhere, because each iteration of Hitfilm has got slower.  Even slapping some text on the screen causes it to skip frames on playback when it's barely doing anything, as in #15 Viewer Playback problem on page 2 of this (blimey it's getting long!) thread.

    ( BTW, Cedric, you might have checked already, but in case you missed it: #25 Dependency Bug? above this one is pretty weird. )

  • NormanPCN
    NormanPCN Posts: 4,082 Enthusiast
    edited October 2016

    @Palacono "As it is, each one takes about an hour to put together, upload and write about"

    I hear you. I've only done a handful, if that, of videos to answer some forum question on the odd chance for some reason I decide to stick my foot in my mouth. Because I don't want to spend the time upping the production value like your vids I just upload the screen capture and hope I was clear. I should get a mic if I ever do a forum answer vid again.

    This "slow" post and the most recent vid answer I did reminded me of something else excruciatingly slow in Hitfilm. Audio waveform display. My CPU is a flat 4Ghz and it hurt waiting for Hitfilm to redraw the waveform when the timeline zoom was changed. Hitfilm is recomputing constantly. Compare that to Vegas that computes it once and redisplays are instant. Sad given that Hitfilm uses it "cache" folder for audio resampling saves. A waveform save could easily go there as well. That would be a better mechanism, IMO, to the Vegas way of leaving the SFK (waveform) files laying around.

    One thing curious in your vid is the CPU utilization. Not the percentage but rather the spread across multiple threads. None of them balls to the wall. I get one thread going balls to the wall and some other housekeeping stuff making the difference. In this example the screen capture running. Your utilization minus the screen capture is about the right % for one thread balls to the wall but the graphs are puzzling. Normally I might ignore this as task manager is not precise but in instances I see Hitfilm not able to get my machine, balls to the wall in CPU or GPU.

    I like the possibility of the timeline recompute. If Hitfilm is not saving results like this then maybe it should. Compute it once instead of every time you hit play. Compute that overhead outside the play sequence. Even a property whose value is a double precision float and 1000 frames in an event only amounts to an 8K array. (33 seconds at 30p) .

  • Palacono
    Palacono Posts: 3,423 Enthusiast
    edited October 2016

    @NormanPCN I suspect that, like many programs, Hitfilm is built on an underlying framework of previous code that, were it written today: would be done differently and more efficiently. 20:20 hindsight is a wonderful thing. ;)

    I've done it myself, where a routine that "does the job" gets more attention than it was ever intended to handle and it's only when you run an optimising scanner over the code that you find the bottleneck. The nice thing is though, that you can end up optimising other code to try and mitigate for the problem before you find it and pull out the cork. Then you're pleasantly surprised by the sudden massive increase in overall speed that you may not have achieved if you'd fixed the original problem first, because then you wouldn't have bothered to fix the other things. For that reason I used to use optimising scanners only occasionally, when I'd run out of other things to fix. They always found something I'd never suspected and of course, every improvement you make pushes the load onto some other part of the program and produces a new bottleneck somewhere else. :D

    There are lots of ways Hitfilm could be optimised, but doing so takes a different approach to adding features (and debugging them). You really want someone who is not the original coder to go through it because you get code blind (and also damn sick of a routine) and are glad to see the back of it once it works as designed.

  • NormanPCN
    NormanPCN Posts: 4,082 Enthusiast
    edited October 2016

    @Palacono "...Hitfilm is built on an underlying framework of previous code that, were it written today: would be done differently..."

    Word!

    For all my development life I have written software development tools and had communication with our customers. Almost without exception nobody ever restructures their dataflow as the needs of the app changes over time or when they realize something could/should be better.. Heck, most developers had no concept of what dataflow design was and that was their real problem.

    I remember being spanked by something that worked fine but when we got a customer that did something HUGE something that performed fine with "normal" stuff and no longer did with that huge stuff. The issue was with a topological sort.

    The amount of tinkering I did, for no other reason other than, this is better for the future, was immense, and it paid off most times. I know more today than I did yesterday and I used that to improve existing implementation. That also made future implementation easier most times. More extensible.

    Always swim downstream. It's the thing a lazy SOB wants to do (aka me). The trick is to route the stream flow so you can swim downstream.

    No programmers really have that freedom. Managers don't normally let it happen even when the programmers have good reason. I had complete freedom being principal.

  • Palacono
    Palacono Posts: 3,423 Enthusiast

    @NormanPCN well, time is money, and programs: like art, are never really finished. At some point the programmer/artist just decides to abandon them so they can be sold. ;)

  • CedricBonnier
    CedricBonnier Posts: 1,197 Staff

    @NormanPCN You've preaching to the choir here. As you said it is sometimes hard to justify for a dev to spend a week or two refactoring something that works (or seems to most of the time), especially when no customer is complaining about it.

    @Palacono Having easily reproduceable steps exposing an issue definitely helps. I have logged #26 in our internal bug tracker. Most of the time a video isn't necessary. Steps to reproduce, eventually a screenshot or a project file is all we use internally. That said, sometimes making a video is quicker than explaining it...

  • Palacono
    Palacono Posts: 3,423 Enthusiast
    edited October 2016

    Well, a video is less subjective than me just saying "I think this isn't working as well as I'd like".

    You can't argue with a video and they're sort of fun to make too.  :)

    @NormanPCN production values aren't really that high (although I did add some subtle Light Rays this time :D ). Having got the first project file, I just do a 'Replace' on the video clip, Change the Title text, adjust the music length to match and export it.

  • Palacono
    Palacono Posts: 3,423 Enthusiast
    edited October 2016

    New Bug/feature: #27 Towards Layer Weirdness
    -------------------------------------

    Not entirely sure about this, but it seems like a bug. :(

    tldr;A layer seems to end up with two parents...

    Original discussion about this was here where @Epitaph_DF was trying to do a 'reverse' HUD effect to get his speech bubble to follow a moving car. 2D to 3D tracking

     We got it working there and then Epitaph_DF implemented it with his own Text Bubble in that thread.



    Just for fun, I then tried also adding something else to the point following the Licence Plate, which would give the same effect as the HUD Tutorial had done.

    But, it didn't work quite as expected, and using different layers to look towards made the angle on the 'HUD Element change', even though both layers it was looking towards were in exactly the same place, because one was parented to the other with zero transform values of its own.

    So why the difference? It's as if the position of the layer that it was looking towards did not take into account the fact that it was parented to something else.

    Edit: Updated the video with a bit more information/experimentation. It seems that the layer that is aligned with Towards Layer is also inheriting that layer's orientation in some way, even though the layer that's doing the looking already has a Parent to inherit information from. So the layer somehow has two parents? How the heck does that work?

    Also, when 'unlooking' and 'relooking' towards another layer the inheritance of the orientation seems to get confused so you end up with an orientation that isn't the same angle as either of the things it was 'Looking towards', or Parented to. It's pretty confusing to try and work out what order you are supposed to do things in  order to be able to understand the expected outcome. :(

    Here's the project file to download and have a play with:
    http://www.mediafire.com/file/burtvy4njnt3cuu/Hitfilm.zip

  • Palacono
    Palacono Posts: 3,423 Enthusiast
    edited October 2016

    New Bug/feature: #28 Alpha Weirdness/Bug
    -------------------------------------

    I've been having some problems with Alpha on some Effects and thought I was doing something wrong, or it was that old Order of Operations 'gotcha' again, but I think I found something else.

    BTW, this applies to all of: HF3P, Hitfilm 3 Plugins, HF4P, HF4E, Hitfilm Ignite.

    Basically, number x zero = 0 and so when number is an Alpha value, no amount of messing about with it should be able to change it, but the video shows where some Effects do anyway; plus going just plain screwy when working with a feathered mask.

    I originally ran into problems when I tried to do Invert Alpha, Matte Cleaner>Choke, Invert Alpha to expand the key on some Green Screen footage. By default: Matte Cleaner only softens the mask inwards, which is no use at all when you're trying to preserve as much of the original image as possible. So the Invert, Choke, Invert would have allowed me to expand the key outwards, then feather it as desired...had it worked. ;)

    Then I was reminded of other times when Alpha behaved screwy on different effects, so put together the examples in the video.

    BTW, the GreenScreen one isn't in there, but that's because it's a subset of the fact that Invert Alpha, Invert Alpha does not get you back where you started, although Invert, Invert does, which works as expected, but is no help with the original problem.

  • Aladdin4d
    Aladdin4d Posts: 2,481 Enthusiast

    While number x zero = 0 is true most of what you're doing here is zero + number = new alpha value or 255 - number = new alpha value

    On top of that the alpha channel always defines transparency values for the entire layer and not just a small area of one so when you make changes to the alpha values you are always affecting the entire layer. 

    Using Invert Alpha multiple times will have a cumulative effect instead of canceling each other out. 

    Mono Halftone affects the area outside the mask the same way Color Halftone does. 

     

     

     

  • Palacono
    Palacono Posts: 3,423 Enthusiast
    edited October 2016

    @Aladdin4d Hmmm..yes, good call. Mono Half tone does affect the area outside the mask if you play with Offset. I'd missed that, as it works with the default values. :)

    However, it at least does have a range where you can rely on the mask, which you can't with the Half Tone Color, so I'm still calling that a bug. :D

    It's actually completely useless for what I wanted to do with it, which was to apply the effect to some text, but when it turns the rest of the screen black it's as much use as a chocolate fireguard.

    Actually, can't ever see a use for it in its current form, other than applying it to the whole screen; as using a mask after the fact (which you can do in an embedded comp) chops right through the blobs, which isn't at all useful, or attractive.

    I'm not sure they should be doing  additions on the Alpha, I'd have thought it would be multiplication (in Alpha Brightness and Contrast)? But I'm open to someone pointing to where these things are 'defined'. Anyway, as Invert+Invert Effect gets back to the original, I'd like  Invert Alpha to work the same way for the Matte, otherwise there is currently no way to do what I want, which is Invert Alpha, Choke, Invert Alpha to expand the mask. 

    So, maybe I need a Really Only Inverts Alpha Effect? :D

    Even if they are doing Addition/Subtraction: 255- original number (0) = 255, then 255-new number (now 255)=0 again, but it doesn't appear to be doing that.

    I've tried stacking other things that mess with Alpha too, but you can  produce an almighty mess like that, especially on semi-transparent Alphas. Rainbows and out of range spots (like the flowers) all over the place.

    I'll do another video tomorrow with Alpha  on a simple square plane; it produces a weird misshapen rainbow pattern that bears no resemblance to the original shape.

  • Aladdin4d
    Aladdin4d Posts: 2,481 Enthusiast

    "It's actually completely useless for what I wanted to do with it, which was to apply the effect to some text, but when it turns the rest of the screen black it's as much use as a chocolate fireguard.

    Actually, can't ever see a use for it in its current form, other than applying it to the whole screen; as using a mask after the fact (which you can do in an embedded comp) chops right through the blobs, which isn't at all useful, or attractive."

    Use an embedded comp for the text and in another comp add a plane with Halftone Color and Set Matte applied in that order using the text comp as the source layer for Set Matte.

    Now I know this is going to be confusing but Inverting Alpha is not the same as inverting a mask or matte like what you want and you are getting getting caught out a little by the order of operations too when you toss in a mask. Here's a project to try:

    https://www.dropbox.com/s/1jfmng5frfthglm/InvertAlphaDemo.hfp?dl=0

    There's two comps. One is a black plane with a mask. The other has a blue plane, a white plane and the black plane comp to use with Set Matte. The Invert Alpha layer has Set Matte and Invert Alpha applied twice. Tick each Invert Alpha effect on to see what it really does. 

  • Palacono
    Palacono Posts: 3,423 Enthusiast
    edited October 2016

    Just a quick response before I'm off to bed. Will try your Project file tomorrow, but I already tried the Set Matte method for Text, it cuts through the blobs, which is not what I want. I want the blobs applied to the Set Matte shape, not cutting it out afterwards. Then you would get nice lumpy text, like you do with the Mono version, but you don't.

    Also  you get exactly the same result as if you just use the Text as a Set Matte source without making it an embedded comp, because the shape of the text is the same as its Alpha layer. TTFN :)

  • Palacono
    Palacono Posts: 3,423 Enthusiast
    edited November 2016

    Well, I worked out how to do what I wanted, which was to expand the mask around a Green Screened actor. :)

    1. Apply the built-in Greenscreen Key Preset and in Matte Cleaner apply 4 pixels of choke and set everything else to zero, then turn on View Matte, Remove the Spill Removal and below that add an Invert. Make this a Composite shot called Green Matte. It should have the character as black against a white background.

    2. In another Comp, put the original Greenscreen footage, the new Background Layer you require below that, then a Grade Layer and finally the Green Matte. Turn off visibility on the Grade Layer and the Green Matte.

    3. Add to the Green Matte a Demult and a Matte Cleaner.

    4. Add to the original Green Screen footage a Set Matte and point it at the Grade Layer above the Green Matte and set the Blend mode to Subtract and add a Spill Removal below that.

    The Green Screen character will now look overchoked (by those 4 pixels applied in the Green Matte Comp), but now the Matte Cleaner>Choke on that bottom Green Matte layer expands the Matte outwards, so that at 4 pixels of choke, it's where zero would normally be. Also Feather expands instead of contracting the Matte. This allows you to choose perhaps one extra pixel or two of Matte around the character, then feather it slightly without eating into the character to do so. If you want more choke to play with, increase the amount in the Green Matte Composite.

    5. For the final cherry, Add a Light Wrap to the top layer of green screen footage and have it use the Background Layer to colour the edges correctly.

    A bit fiddly to do and ideally: Matte Cleaner being able to choke and feather in both directions like a Mask's Shape can - instead of only one - would mean it could all be done in a single Effect.

    Either that, or Invert Alpha actually inverting the Alpha - I can't work out what it's currently doing - then a couple of those either side of the Matte Cleaner, as I tried originally, would also work.

    But I suspect, as is usually the case, I'm trying to hit the hammer with the wrong end of the chisel. :D

  • Palacono
    Palacono Posts: 3,423 Enthusiast
    edited November 2016

    New Bug/feature: #29 Blurry Patterns
    -------------------------------------

    Having played around a bit more with Alpha - the Histogram is a great way to see what's going on as well as being able to adjust everything - it looks like it is impossible to create a simple Invert Matte equivalent, although you can get fairly close. But then you find that stacking Effects that play with Alpha just leads to madness; Matte Cleaner>Choke acting like Brightness?

    But as well as most of my last video probably being able to be classified under 'Misunderstandings', there is still a weird 'out of range' bug  in there and this in a similar vein.

    But, it is the very definition of an edge case and not something you'll likely encounter in normal use, TBH. ;)

  • Palacono
    Palacono Posts: 3,423 Enthusiast

    New Bug/feature: #30 Invisible Keyframes Bug
    -------------------------------------

    It's not the keyframes themselves that are invisible - you can see those - it's what they do to what's on the layer: namely some text in this particular instance; which instead of just moving a few pixels over 5 frames, goes invisible, then reappears on the next keyframe.

    Clue/Red Herring for Devs?: Not shown in the video for simplicity, but if I turn Motion Blur on for the layer, the text goes semi-transparent for one frame before it goes completely invisible on the next, and stays that way. So it might be that something thinks the text is moving really, really fast; too fast to see? ;)

    Here's a link to the project file to play with.
    Invisible_text.hfp Project File

  • CedricBonnier
    CedricBonnier Posts: 1,197 Staff

    @Palacono #30 is a bug in HitFilm (2-4). Your text layer disappears because it was 3D and contains Z data. To fix this, either set your text layer to 3D, delete (set to 0) the Z value of the keyframe and set it back to 2D or simply delete the keyframe and create it again.

    Sorry about the inconvenience.

  • Palacono
    Palacono Posts: 3,423 Enthusiast
    edited November 2016

    @CedricBonnier OK, no problem. I don't think I'd ever made the layer 3D when messing about, although there were originally some scale values and possibly some effects, but I removed those else until just the single set of Transform values remained and it still did it.

    If it might help with debugging: I think it went adrift after I'd used ALT to close up the gaps between the points, as they were originally more spread out. BTW that works a bit weirdly. If you don't have a keyframe at frame zero, it's often very difficult to get it to scale them - it moves them instead. Drop an extra keyframe in at frame zero and scale works. Mostly.

    I know in the past (HF3?) ALT-scaling produced sub-keyframes that were greyed out and you couldn't actually move to them, but I think that was fixed for HF4? Although you still get grey keyframes when tracking something, even when everything is at the same framerate, as mentioned as an aside in the #18 Quad Warp Problem Deja Vu video.

  • Palacono
    Palacono Posts: 3,423 Enthusiast
    edited November 2016

    New Bug/feature: #31 Export Speed Extra Digits
    -------------------------------------

    Just a quickie that might even have been fixed in HFP2017 (not installed yet to check) but those with HF4P and HF4E should hopefully have a few updates due before they both become EOL in a few months, so I better get a move on with even the niggly ones.

    When Exporting 60fps (59.94) video (and I think also when creating a project that matches the footage when you drag it into the Editor first, forgot to check, but have a vague memory of that) the speed is shown as 59.9401 fps and the line around the box is a bit wonky as the text cuts into the bottom of it.

    Setting it to 59.94 manually seems to stick and AFAICT makes no difference when exporting, as a small 14 second clip was exactly the same length with both 'speeds'. If it did make a difference it probably wouldn't show up until the video was hours long. :)

    http://i854.photobucket.com/albums/ab106/pickaname2/speedbug.jpg
    http://i854.photobucket.com/albums/ab106/pickaname2/speedbug2.jpg
    http://i854.photobucket.com/albums/ab106/pickaname2/Speed Compare.jpg

     

  • Palacono
    Palacono Posts: 3,423 Enthusiast
    edited November 2016

    New Bug/feature: #32 HF4 Tracking Semi-Keyframes
    -------------------------------------

    This is an oldie for HF4P and HF4E - I referenced it back in #18 and we've had updates since then but it's still there.  Doesn't seem to have any particularly detrimental effect, just looks a little off, but might be light relief for a developer from the bigger bugs being found in HFP2017.  ;)

    Oh yes: the Project settings, the clip properties and everything else are all set to the correct frame rate, so it's not that.

    Next video will be on a more serious proxy bug that I reported back in June, but is still there and makes using proxy with some images unusable, especially environment maps. Will upload a project file so people can check it out on HFP2017.

  • CedricBonnier
    CedricBonnier Posts: 1,197 Staff

    @Palacono what you see in #32 is keyframes that do not land perfectly on a frame. This by itself isn't a bug as it is valid to have sub frame keyframes (this will happen for example when you scale keyframes). Even if you cannot navigate to it, the keyframe still works and affects the property as expected. You can fix this by zooming the timeline all the way in and move the keyframe so that it lands perfectly on a frame.

    I understand that even if it works, seeing these sub frame keyframes is slightly annoying, I'll log this in our internal bug tracker.

  • Palacono
    Palacono Posts: 3,423 Enthusiast
    edited November 2016

    @CedricBonnier Not a bug? :) No scaling was applied, just track, and either Stabilise or Transform. No other user intervention. One works, one gets it ever so slightly wrong. But as minor as the 59.9401 fps on Export, so barely a buglet. ;)

    But, just discovered a frustrating one with the Camera in HF4. Set a keyframe with the orientation at 0,0,0, then set another one a few frames later also at 0,0,0 (basically hold position for a while before moving on) and the camera instead cycles through 360 degrees on Y between the two points. I'll put together a video for that because something else weird popped up while messing with that when trying to move something in 3D with two view windows open. They might be interdependent.

    BTW, a previous camera bug was fixed in HF4 when importing .ma files from various programs that caused the camera to flip out when crossing an angle boundary at 0 with the result that the camera just didn't draw anything for that frame. I reported this for HF3P ages ago and was glad when I recently found it started working in HF4 for all the MatchMoving programs that I thought were exporting buggy .ma files when I tried it originally. Wish I'd tried it sooner in HF4E, but I thought the other programs were at fault and the fix wasn't documented in the HF4 release notes and I can't import 3D models anyway.

    I have HF3P, and HF4E only. If I import the .ma file into HF3P the camera gets upset and throws frames away. I then save the project. I do the same for HF4E (or HF4P) and it gets it right and I save the project.

    If I import both saved projects into HF4E (or HF4P) the original error remains in the HF3P project and it flips out the camera, so the fix is apparently being done on the original import of the .ma files into HF4 and is not an inherent fix on how the camera works in HF4.

    However, I want to use HF3P because it mostly does what I need and also because I can't use 3D models in HF4E, only 3D Planes. But I can't load even the simplest of HF4 files (such as one with the corrected camera data) into HF3P because of the version differences. So I'm stuck.

    It's pretty frustrating to not be able to find a cheap/free program (maybe the expensive ones work? No idea) that can export .ma files in a format that HF3P will accept, so I can't make use of the feature at all.

    Polite Request
    ===========
    Would it be possible for FXHome to release either a teeny tiny standalone program that can convert the .ma files into something that HF3P can use (the values really do change on import to HF4, so load/save imported .ma data immediately would appear to be possible) Or for some projects (really, really simple ones with no Effects, nothing that could break things ) to be able to be exported from HF4 in HF3 compatible format so they could be loaded with no "version error"?

    I know that HF3P is EOL and I'm stuck with lots of other relatively minor bugs that won't ever get fixed, and you'd prefer I upgrade to get the fixes, but that one limits things quite a lot. Basically no Matchmoving for me. :(

  • Palacono
    Palacono Posts: 3,423 Enthusiast
    edited November 2016

    Messing with Motion Blur produced some weird results.

    Didn't initially seem that buggy until I experimented more... when it did,  so linking it here:

    https://hitfilm.com/forum/discussion/41990/how-do-i-fake-or-force-motion-blur-hf4

  • CedricBonnier
    CedricBonnier Posts: 1,197 Staff

    @Palacono just to be clear, I didn't say #32 wasn't a bug, it is. I said that in general, sub frame keyframes are not a bug and there are legitimate cases where this happens (eg scaling keyframes, changing timeline framerate, etc).

    As I said, the sub frame keyframes do work. The fact that when copying them to a point layer changes the time is a bug (I've logged it internally) but if the keyframe is 1 or 2 ms off the intended frame is unlikely to mess your tracking.

  • Palacono
    Palacono Posts: 3,423 Enthusiast

    @CedricBonnier Oh, OK. I reparsed your original response and see that's what you meant now. :)

    Confused me slightly as it's a bit like crashing into someone's car and saying: "Those dents are caused by the impact, which is what you'd expect when two cars collide..." :D

    Anyway, about that little .ma file fixer for HF3P customers, any chance of that, or are all HF3P owners SOL unless they upgrade?