future wish : AI based video slow motion
There is a new trend in video editing where neural networks have learned to slow down video, so a 25fps can do time stretch (slow down) withouth strange blur or jitter effects a Nvidea demo from 2 minute papers is shown here youtu.be/LBezOcnNJ68
I think it would be great for matching video to animation speeds in vfx to be able to stretch / compress time of original video mpg / avi files
2 Comments
well i wouldnt mind if vse will have something like dain github.com/baowenbo/DAIN / grisk.itch.io/dain-app (but it works only with cuda) integrated or this for deep compositing roxanneluo.github.io/Consistent-Video-Depth-Estimation/ with this youtu.be/nCpGStnayHk (could be very good combination for post processing video and making vfx work easier/"faster"-because manual rotoscoping is much slower/and more accurate) microsoft.com/en-us/research/publication/unwrap-mosaics-new-representation-video-editing/ or something like this for fast video editing - but then blender will need also some inpainting features for texturing and video editing i think. Also rotoscoping could be more handy: like rotodraw and rotosnap youtu.be/K5Y3GNJ4r7E (its like magnetic lasso in photoshop/affinity photo) youtu.be/L1JaJjprzmA also mocha pro has now this and silhouettefx: youtu.be/NwbbHFO8Rl0 (these reminds a little bit from gimp editors foreground selection tool) and megaplates (extracting video into panorama like that microsoft's mosaic unwrap): youtu.be/V0pRmGGmdT4 also this is pretty awesome: aescripts.com/lockdown/ or content-aware/inpainting for vfx work youtu.be/25ltIoHtiO4 or deepfake (but with opencv / face tracing and transferring that data for 3d rigs) - at the moment there is just adobe's character animator and cartoon-animator (few ohters too) for 2d face tracings and animations - grease pencil would be very cool with opencv face capturing :) like this youtu.be/RY_eErKlilw but with 2d rigs
So much things that are cool and very welcome to blender if these will come into blender.
But i agree - that slow motion (neural network optimized optical flow with depth aware would be very lovely feature) Sorry for long post.
* @juhamukari * i follow a lot of those topics you mentioned as well :) We can agree there are a lot of nice things out there. With this i was also thinking the ease of slowing down smoke or fire simulations without aritfacts.