Optical Flow MEMFOF Tartan T TSKH

🚀 MEMFOF-Tartan-T-TSKH

MEMFOF is a memory-efficient optical flow method designed for Full HD video. It combines high accuracy with low VRAM usage. This specific checkpoint is tailored for real - world videos, trained with higher diversity and robustness in mind.

📄 Paper | 🌐 Project Page | 💻 Code | 🚀 Colab | 🤗 Demo

🚀 Quick Start

🔍 MEMFOF offers a memory - efficient solution for optical flow estimation in Full HD videos, balancing high accuracy and low VRAM consumption. This checkpoint is optimized for real - world video scenarios.

✨ Features

• Memory - efficient: Ideal for Full HD video with low VRAM usage.
• High accuracy: Delivers precise optical flow estimation.
• Robust for real - world videos: Trained with high diversity and robustness.

📦 Installation

git clone https://github.com/msu-video-group/memfof.git
cd memfof
pip3 install -r requirements.txt

💻 Usage Examples

Basic Usage

import torch
from core.memfof import MEMFOF

device = "cuda" if torch.cuda.is_available() else "cpu"
model = MEMFOF.from_pretrained("egorchistov/optical-flow-MEMFOF-Tartan-T-TSKH").eval().to(device)

with torch.inference_mode():
    example_input = torch.randint(0, 256, [1, 3, 3, 1080, 1920], device=device)  # [B=1, T=3, C=3, H=1080, W=1920]
    backward_flow, forward_flow = model(example_input)["flow"][-1].unbind(dim=1)  # [B=1, C=2, H=1080, W=1920]

📚 Documentation

Citation

@article{bargatin2025memfof,
  title={MEMFOF: High-Resolution Training for Memory-Efficient Multi-Frame Optical Flow Estimation},
  author={Bargatin, Vladislav and Chistov, Egor and Yakovenko, Alexander and Vatolin, Dmitriy},
  journal={arXiv preprint arXiv:2506.23151},
  year={2025}
}

📄 License

This project is licensed under the BSD 3 - Clause license.