hzwer / Practical-RIFE

More practical frame interpolation approach.
MIT License
625 stars 68 forks source link

How to solve this problem? #83

Open lionessbzzz opened 4 months ago

lionessbzzz commented 4 months ago

Occurs in versions starting from 4.18, on older versions everything works fine

22

TNTwise commented 4 months ago

Isn't that flowframes? Flowframes doesn't support newer models. You can try ncnn with the binary swapped and that should work.

lionessbzzz commented 4 months ago

Isn't that flowframes? Flowframes doesn't support newer models. You can try ncnn with the binary swapped and that should work.

If it's not too much trouble, can you show me in pictures how to do this? (I'm writing through a translator) And what are the current alternatives to flowframes that have normal deduplication (for processing anime) And in general, flowframes are dead?

TNTwise commented 4 months ago

Flowframes is mostly dead. He is apparently going to update it, but it still doesn't include the latest models. While I wouldn't know how to swap the models step by step, you might be able to use something like https://github.com/NevermindNilas/TheAnimeScripter. If you want to use flowframes, find the vapoursynth rife ncnn within flowframes directory (may be called librife.dll) and you can change out the models with models from https://github.com/TNTwise/rife-ncnn-vulkan

Ajaja commented 3 months ago

I use rife-ncnn-vs. So, I've downloaded models and the latest FlowframesData\pkgs\rife-ncnn-vs\vapoursynth64\plugins\RIFE.dll from https://github.com/styler00dollar/VapourSynth-RIFE-ncnn-Vulkan , got the latest flowframes code from https://github.com/n00mkrad/flowframes and compiled it, then added 4.20 to FlowframesData\pkgs\rife-ncnn-vs\models.json and files.json to FlowframesData\pkgs\rife-ncnn-vs\rife-v4.20\files.json (rife-v4.20 folder is rife-v4.20_ensembleFalse from VapourSynth-RIFE-ncnn-Vulkan) :

[
  {
    "filename": "flownet.bin",
    "dir": "",
    "size": 30826032,
    "crc32": "3335961018"
  },
  {
    "filename": "flownet.param",
    "dir": "",
    "size": 28654,
    "crc32": "3220769076"
  }
]

It works. But 4.20 model is ~2x slower than 4.6 in Flowframes v1.40.0 :(