I have a question/maybe request for a python script. It could be a nice tool to supply a TimeWarp to the MLtimewrp.
Question is, is python something that if it is written already can it be “easily” adapted.
I saw this for nuke and wondered if this could be adopted to flame?
I don’t know python at all but this kind of stuff really makes me want to learn it so that I might be able to help build tools that will help us all.
If someone gets an itch or knows someone that could make this or something like this work in flame it could be a real time saver.
I do not believe so. That tool lets you say what frames are bad ie too much motion blur, camera bump, flash frame or something over multiple frames and single frames for a whole shot length. I do not think the current ML timewarp has a way to know what frames you want to remove and rebuild.
If some rad person makes this we could feed that timewarp into the ML one for state of the art repaired frames.
Couldn’t you take the clip with bad frames, remove them, and then slow that down with a flame timewarp to the original destination duration that feeds the ML?
It’s not as simple as the original Furnace with that Nuke tool is based on, me thinks, where you used to specific bad frames, but, the Machine Learning from what I know creates in betweens, which, sounds similar but in practice is quite different.
I just tried the fill/remove option by holding good frames in a timewarp rendered it and then fill/removed that result with interpolate. Very nice only new frames show up on a difference matte check. Disregard the request. Weekend is back on!!!
TimewarpML can do it at the moment already it is just a good interface for batch is missing. Just repeat the frame before and place it over the bad frame (or several frames) and then run it through Interpolate/Remove tool set to “Interpolate” - it will do the job
If you’re able to send the example over I can probably show it tomorrow on Logik live. If not I will try to make something up but real example is easier
Hey Talsoh,
Sorry for the late reply. I had quite the honey do list today. I have a set up I can try and sent to you. Might be too late.
But I think what from I am seeing the removing of the bad frames and time warping them back in seem to give better results.
Let me know or maybe a moderator can help out.
Cheers.
Hi Brian, there was a demo of this functionality yesterday (with “made-up” damaged footage), it is at around 13:15 in the recording, let me know if it helps
Thanks Andriy, I watch it yesterday. I was wondering while watching if the artifacts toward the top of the temple could be further reduced by letting the ML model know what frame is closet to the correct one with the timewarp method I am talking about in this thread.
I saved a batch setup with a similar “made-up” bad frames and all the different ways to repair them. I can send it you if you would like to take a look.
It looked like the removing the bad frames and time warping the correct frames back into their original timeline positions with the ML timewarp gave better results on multiple frame patching than just holding a single good frame for several frames and using the removing duplicate frames when I differenced them to the original non “made-up” bad frames clip.
Hope that is making sense.
I am so blown away by this tool. Thank you again for all the work you are putting into it.
Hiya @ihatebrianwilliams do you mind if I join this thread with the original ML Timewarp thread? This would be really great to have it all together for future sake.