Modelpublic

a-yakovenko/FreeFlow-L-pretrain

sourceHugging Facecc-by-nc-sa-4.0updated 13d agoView on Hugging Face

0likes60downloads

Model Card

FreeFlow-L · pretrain

Cross-view pretraining · 211,677,376 parameters

Code | arXiv

Setup

Clone the code repository and install the package (Python ≥3.10, PyTorch ≥2.5):

bash

git clone https://github.com/msu-video-group/freeflow.git
cd freeflow
python -m pip install -e ".[examples]"

See the README for uv/Conda setup and CUDA installation. The examples below use a CUDA-enabled PyTorch build.

Usage

python

import torch
from freeflow import FreeFlowPretraining, normalize_images

model = FreeFlowPretraining.from_pretrained("a-yakovenko/FreeFlow-L-pretrain").eval().cuda()
# Two RGB views, resized to 224 × 224; random pixels for a minimal example.
images = normalize_images(torch.randint(0, 256, (1, 2, 3, 224, 224))).cuda()
with torch.inference_mode(), torch.autocast("cuda", dtype=torch.bfloat16):
    reconstruction, mask, target = model(images[:, 0], images[:, 1])

Returns reconstructed 8×8 patches, a sampled mask, and target patches. This checkpoint does not estimate optical flow.

Citation

bibtex

@inproceedings{bargatin2026freeflow,
  title={FreeFlow: A Bias-free Hierarchical Transformer for Optical Flow Estimation},
  author={Bargatin, Vladislav and Yakovenko, Alexander and Abud, Khaled and Vatolin, Dmitriy},
  booktitle={European Conference on Computer Vision (ECCV)},
  year={2026}
}