I need to covert some HD interlaced (25i and 29.97i, aka 50i and 59.94i) footage to UHD 50p.
I’m in a UHD 50p project and using Batch FX with deinterlace which separates my fields and plays back the video at half speed, then timewarp to bring the footage back to the correct speed again.
This gives good results in the temporal domain without having to use any motion estimation. There’s no frame doubling and each field has become a full frame.
I am now seeing aliasing-type artefacts because every frame is missing half a line.
Are there any clever tricks I can use to do a decent estimation of what should be in these missing lines, and also handle the upscale
from HD to UHD?
Has anyone ever done this conversion before? It’s quite an unusual one!
I have VideoEnhanceAI and haven’t used it on interlaced footage but I know there’s an option to read it. I just did a quick search and apparently there’s a model that automatically deinterlaces (VE-AI uses different models so is flexible as to what algorithm you use) but the person said they were getting artefacts on that model. Perhaps you could use the VideoEnhance and then deinterlace in Flame afterwards? I know there’s a demo that’s watermarked so you could give it a go? Before I bought it there was a full timelimited version, but they went the watermark route. Still you’ll see if it works and then it might be worth investing in as it has many uses and it’s great for seeing to old video you’d perhaps given up on?
Cheers
Tony
*Would love, btw, for Autodesk to buy up this amazing software and make it into a node that you could mask and matte around, etc. I am not sure how prized this software is in their stable as I don’t see it being updated very often, and I get emails on their other software but not on this. Maybe they’d sell it for the right price Autodesk and you could integrate it like Blackmagic seem to be able to do so easily with everything they buy, in like two weeks after purchase! I remember Will Harris mentioning resize in a presentation on Machine Learning in Flame but it hasn’t happened so far, sadly. This would be like a fantastic resize node on steroids. Get an AI batch timewarp and AI resize and carry on expanding and improving AI/machine learning stuff.
Soz, I’m too lazy to do this properly and to replace the original rererence to “Rooney” with the proper look, but those of a certain age and from the pUKe, and who are perhaps ManUtd fans, might just feel a certain connection with ManUtd and Flame and of wanting “The Glazers” to buy up players to help to grow and improve the team? Rooooooooney…
Topaz was likely built to deinterlace and change the size of moving images, but not necessarily at the same time, and likely has no built in mechanism for framerate conversions as that process in and of itself is pretty specialized and reserved for um…specialists…like us.
What I should have stated earlier instead of just dropping a link was that I think you should break this process down into a few steps.
First, use Topaz to deinterlace the footage. That ought to be step 1. If there is a built in deinterlacer and upscaler concurrently, then awesome. I’ve personally fed it lots of images but nothing explicitly interlaced so my memory is a bit fuzzy in that regard. Step 2 get it the size you want. Then, bring that into Flame and try a Machine Learning Timewarp set to 1/2, then change THAT to 50fps.
That’s likely using each bit of kit in its strongest area.
