Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!
Blackfyre, of course not and there is a good reason for this. The RIFE models that are described in that paper you linked to: RIFE, RIFE-Large and RIFEm are optimised to achieve the best possible PSNR and SSIM. Unfortunately, many researchers stop there, because that is enough for publications and citations. However, there are researchers who also provide models optimised for human perception and measures that better reflect it: LPIPS, FloLPIPS, VFIPS.
RIFE developer hzwer did not include models optimised for human perception in the paper you linked to, but he did in a separate repository to which you have a link here: https://github.com/hzwer/Practical-RIFE The practical models are, of course, based on the models contained in the scientific paper you provided a link to and so, for example, the v4.0-v4.6 practical models are based on the RIFEm model. The difference is, the v4.0-v4.6 practical models are better optimised in terms of human perception, inference performance and trained on more data. These models use SVP, including the latest v4.6, but not directly.
To further improve performance and take advantage of the fastest Tensor cores of NVIDIA graphics cards for AI inference, SVP uses versions of these models from this repository: https://github.com/AmusementClub/vs-mlrt The authors of this repository have recently included a second version of the implementation of these models, and described the changes in this update: https://github.com/AmusementClub/vs-mlr … c649dfb212
To summarise:
The best current model is:
rife_v4.6_ensemble.onnx from the v2 package: https://github.com/AmusementClub/vs-mlr … e_v2_v1.7z
If someone is not able to use the full capacity of their screen with this model, for example to interpolate 4K files, it is worth using a faster model, with a small loss of quality:
rife_v4.6.onnx from the v2 package: https://github.com/AmusementClub/vs-mlr … e_v2_v1.7z
If anyone is interested in much, much better quality than RIFE interpolation I will write about it soon, here on this forum. Please give me another week, maybe two weeks, until I update my repository on GitHub.