gtomato/practical-rife-v4.25

Practical-RIFE v4.25 — flownet.pkl checkpoint

Mirror of the official Practical-RIFE v4.25 interpolation checkpoint (flownet.pkl) for reproducible, checksum-verified downloads. Originally distributed by the upstream author (hzwer) as RIFEv4.25_0919.zip on Google Drive; mirrored here so it can be fetched with standard Hugging Face tooling.

Contents

File	Description	SHA-256
flownet.pkl	IFNet_HDv3 weights (PyTorch state_dict)	6615790efd627772917205db291f51cd392528a157ecbb2ecaeec3bff8eb6de2

The inference source is not included here. The consuming project integrates it directly as rife_ifnet.py and rife_warplayer.py; see provenance.json for the upstream origin record.

Provenance

• Upstream repository: https://github.com/hzwer/Practical-RIFE
• Upstream version: 4.25
• Source commit (for the vendored modules): 17d8c7a1005b37f4c97bfee04e316aaec7fdc536
• Checkpoint archive: RIFEv4.25_0919.zip (sha256 e63d481b7ae5d4a4e6ad7ac5b410ff78f3bf7be3b51b2e38ca8152747abde5b4)
• License: MIT (see LICENSE)

Usage

Load with PyTorch and feed into the IFNet_HDv3 architecture:

import torch
raw = torch.load("flownet.pkl", map_location="cpu", weights_only=True)
active = {k.removeprefix("module."): v for k, v in raw.items()
          if not k.removeprefix("module.").startswith(("teacher.", "caltime."))}

License

MIT — see LICENSE. Copyright (c) 2021 hzwer.