GIMM-VFI (fastforks)
GIMM-VFI (Guo et al., NeurIPS 2024), the RAFT-based GIMM-VFI-R: arbitrary-time video frame interpolation. RAFT estimates the flow between both frames in both directions; a generalizable implicit motion model (a SIREN MLP on time and position, conditioned on the forward-splatted flow latents) predicts the flow at the requested time; an AMT-style decoder refines it with the bidirectional feature correlation and blends both warped frames. Any time in [0, 1]. gimm-vfi-r is trained for accuracy, gimm-vfi-r-p adds an LPIPS loss for sharper, perceptually better frames.
This repo is the model card only: it holds no weights. fastforks downloads the authors' weights from the links below and checks them against the SHA-256 in the model card. Code: https://github.com/kleinicke/fastforks. Kernels for the kernels backend: https://huggingface.co/fastforks/kernels. Original code: https://github.com/GSeanCDAT/GIMM-VFI at dbc56449994a.
Author weight repos: https://huggingface.co/GSean/GIMM-VFI.
Task: interpolation. Tier: experimental. Outputs: frame. Backends: portable, kernels.
Input: two RGB images (consecutive frames). Output: the frame between them at timestep (0.5 by default; 0 = the first, 1 = the second), RGB in [0, 1].
Variants
Weights are the authors' files (not hosted here). SHA-256 shortened to 12 hex digits; the full hashes are in card.toml.
| variant | weights (author link) | SHA-256 | weights licence | commercial use |
|---|---|---|---|---|
gimm-vfi-r@1 |
gimmvfi_r_arb.pt | 1dff0ac3ca91 |
LicenseRef-noncommercial | no |
gimm-vfi-r-p@1 |
gimmvfi_r_arb_lpips.pt | b33e41536850 |
LicenseRef-noncommercial | no |
Licence
Original-code strings, recorded as they are (research-use package; check them before any other use). Code: LicenseRef-noncommercial. Licence files: LICENSE.
Training data: Vimeo-90K (septuplets / triplets for frame interpolation) (configs/gimmvfi/gimmvfi_r_arb.yaml: Vimeo-90K septuplets with arbitrary-time targets; the RAFT flow estimator starts from the raft-things checkpoint (FlyingChairs + FlyingThings3D).).
Licence evidence: https://github.com/GSeanCDAT/GIMM-VFI/blob/dbc56449994a3c2e045d46fd46bed570239913f8/LICENSE ; https://github.com/GSeanCDAT/GIMM-VFI/blob/dbc56449994a3c2e045d46fd46bed570239913f8/README.md (weights: https://huggingface.co/GSean/GIMM-VFI)
Input size
any H x W >= 128, padded to a multiple of 32
Tested up to (portable, fp16, RTX 5060 Ti 16 GB): gimm-vfi-r 1024x2048 (13.4 GiB peak allocated, 14.5 reserved); gimm-vfi-r-p 1024x2048 (13.0 GiB peak allocated, 14.2 reserved)
Speed and memory
Interpolation, synthetic, 1080x1920
| model | time (ms): original | portable | kernels | kernels speed-up: vs original | vs portable | peak memory (MiB, allocated / reserved): original | portable | kernels |
|---|---|---|---|---|---|---|---|---|
gimm-vfi-r |
OOM | 2622 | 1629 | β | 1.6x | β | 9899 / 10400 | 4587 / 5362 |
gimm-vfi-r-p |
OOM | 2622 | 1628 | β | 1.6x | β | 9899 / 10400 | 4587 / 5282 |
Startup and first call
| model | load (s): original | portable | kernels | first call (s): original | portable | kernels |
|---|---|---|---|---|---|---|
gimm-vfi-r |
2.05 (480x854) | 0.27 | 0.59 | 1.34 (480x854) | 2.62 | 1.63 |
gimm-vfi-r-p |
2.96 (480x854) | 0.27 (480x854) | 0.58 (480x854) | 1.44 (480x854) | 1.00 (480x854) | 0.68 (480x854) |
torch.compile and TensorRT
| model | input | time: portable | compile | TensorRT fp16 | kernels | preparation: compile (cold first call) | ONNX + engine build | peak MiB (alloc / reserved): portable | compile | kernels | TensorRT (engine + torch) |
|---|---|---|---|---|---|---|---|---|---|---|---|
gimm-vfi-r |
1080x1920 | 2622 ms | deferred (build >5 min) | deferred (build >5 min) | 1629 ms | β | β | 9899 / 10400 | β | 4587 / 5362 | β |
gimm-vfi-r-p |
1080x1920 | 2622 ms | deferred (build >5 min) | deferred (build >5 min) | 1628 ms | β | β | 9899 / 10400 | β | 4587 / 5282 | β |
Fixed-size cells: time (ms) / peak allocated / peak reserved memory (MiB).
Hardware, protocol and caveats: benchmarks (RTX 5060 Ti 16 GB, sm_120). Original = the original code named above at the precision of portable (fp16, or (fp32) where fp16 is unsupported or not finite); original fp32 = its default fp32 run. Peak memory is the torch allocator's.
Accuracy
On public datasets
DAVIS 2017 trainval 480p (JPEG), frames 0 / 2 and 20 / 22 of each of the 90 videos give the frame 1 / 21 at time 0.5, compared with the real one: PSNR and SSIM (Gaussian 11 x 11, sigma 1.5) of the output rounded to uint8, on RGB in [0, 1], averaged over the triplets. Original = the original code named under Details, same checkpoint and protocol.
| variant | dataset | original fp32 | original fp16 | portable | kernels | note |
|---|---|---|---|---|---|---|
gimm-vfi-r |
DAVIS 2017 480p, 2 frame triplets per video (180), PSNR / SSIM | 28.94 / 0.8543 | 28.94 / 0.8543 | 28.94 / 0.8543 | 28.94 / 0.8543 | |
gimm-vfi-r-p |
DAVIS 2017 480p, 2 frame triplets per video (180), PSNR / SSIM | 28.81 / 0.8481 | 28.81 / 0.8481 | 28.81 / 0.8481 | 28.81 / 0.8481 |
Measured on NVIDIA GeForce RTX 5060 Ti, torch 2.14.1+cu130, fastforks e97998ab, 2026-11-14.
Parity with the original code
Every variant is checked against reference outputs of the original code at its pinned commit. CPU, float32: max |diff| <= 5e-05 vs the original. GPU, float16 (the default), portable and kernels: mean |diff| <= 0.0003 and p99 <= 0.003 (absolute error of the output frame).
Usage
from fastforks import interpolation
# non-commercial weights: run with FASTFORKS_NONCOMMERCIAL=1 in the environment
model = interpolation.load("gimm-vfi-r") # mode="auto": the preferred backend from recorded measurements
result = model("frame1.png", "frame2.png") # timestep=0.5: the middle frame
result.frame # [3, H, W] float32 RGB in [0, 1], input resolution
result.backend # what ran, e.g. Backend(mode='portable', device='cuda:0', dtype='float16')
Install with pip install fastforks (see https://github.com/kleinicke/fastforks). Other variants: gimm-vfi-r, gimm-vfi-r-p.
Kernels backend
interpolation.load("gimm-vfi-r", mode="kernels") runs our CUDA kernels (raft_flow, cudnn_hacc). It needs a prebuilt kernel for your GPU architecture (downloaded from https://huggingface.co/fastforks/kernels on first use) and a recent NVIDIA driver; if there is none it raises BackendUnavailableError with the reason and never falls back silently. mode="auto" picks kernels when it can run and logs the choice.