Stabilizing Streaming Video Geometry via Dynamic Feature Normalization

---

DyFN improves temporal stability for streaming video geometry estimation while preserving competitive depth accuracy across diverse benchmarks.

🛠️ Installation

Install via pip

pip install git+https://github.com/shawLyu/Streaming_DyFN.git

Or clone this repository locally

git clone https://github.com/shawLyu/Streaming_DyFN.git
cd Streaming_DyFN

Then install dependencies:

pip install -r requirements.txt

DyFN is compatible with a wide range of dependency versions. Check requirements.txt if you need strict environment control.

🎬 Inference

Run video inference with:

python moge/scripts/infer_video.py --video_path videos/jump.mp4 --save_video

🧠 Pretrained Models

Our pretrained models are organized as follows:

Version	Checkpoint	Base Model	Video Stability	#Params
DyFN	shawlyu/DyFN / ./pretrained/DyFN.pt	MoGE-based	✅	320M

NOTE: More checkpoint variants will be released soon.

📊 Evaluation

Detailed evaluation instructions are available in docs/eval.md.

Run baseline video evaluation with:

python moge/scripts/eval_video_baseline.py --video_dir_path ~/data_disk/dataset/local/depthcrafter/datasets/ --pretrained shawlyu/DyFN

Video Evaluation

Method	Sintel (50 frames) Abs Rel↓	Sintel (50 frames) $\delta < 1.25$ ↑	Scannet (90 frames) Abs Rel↓	Scannet (90 frames) $\delta < 1.25$ ↑	KITTI (110 frames) Abs Rel↓	KITTI (110 frames) $\delta < 1.25$ ↑	Bonn (110 frames) Abs Rel↓	Bonn (110 frames) $\delta < 1.25$ ↑
Marigold	0.532	51.5	0.166	76.9	0.149	79.6	0.091	93.1
DAV1	0.325	56.4	0.130	83.8	0.142	80.3	0.078	93.9
DAV2	0.367	55.4	0.135	82.2	0.140	80.4	0.106	92.1
MoGe v1	0.216	65.3	0.117	84.7	0.076	96.0	0.074	95.5
DepthPro	0.319	52.0	(0.088)	(92.7)	(0.088)	(92.2)	(0.063)	(96.6)
MoGe v2	0.214	69.5	(0.110)	(88.2)	(0.183)	(58.8)	(0.049)	(98.0)
VGGT	0.287	66.1	0.031	98.5	0.070	96.5	0.055	97.1
Monst3R	0.335	58.5	0.123	83.2	0.104	89.5	0.063	96.4
CUT3R	0.421	47.9	0.097	88.7	0.118	88.1	0.078	93.7
TTT3R	0.404	50.0	0.114	87.7	0.113	90.4	0.068	95.4
DepthCrafter	0.270	69.7	0.123	85.6	0.104	89.6	0.071	97.2
VDA	0.300	63.3	0.075	95.4	0.079	95.0	0.051	98.1
FlashDepth	0.265	64.2	0.101	90.3	0.103	89.5	0.053	98.0
Ours	0.180	73.0	0.073	96.6	0.062	97.3	0.044	98.4

NOTE: Values in parentheses () indicate evaluation on raw metric outputs without alignment.

Image Evaluation

Method	Sintel Abs Rel ↓	Sintel $\delta < 1.25$ ↑	Scannet Abs Rel ↓	Scannet $\delta < 1.25$ ↑	KITTI Abs Rel ↓	KITTI $\delta < 1.25$ ↑	Bonn Abs Rel ↓	Bonn $\delta < 1.25$ ↑
DAV2	0.200	74.1	0.039	98.2	0.073	95.3	0.048	98.0
MoGe v1	0.124	83.7	0.027	98.6	0.044	98.0	0.028	98.8
CUT3R	0.428	55.4	0.064	93.7	0.092	91.3	0.063	96.2
VDA	0.200	75.3	0.041	98.1	0.074	95.1	0.039	98.6
FlashDepth	0.174	75.6	0.056	96.3	0.085	92.6	0.043	98.7
Ours	0.124	83.7	0.027	98.6	0.044	98.0	0.028	98.8

🏋️ Training

Training and finetuning instructions are available in docs/train.md.

🗂️ Data Processing

Detailed data preparation and processing instructions are available in docs/train.md.

🤝 Acknowledgement

Our code is built heavily on top of MoGE, and the main part of this repository comes from the MoGE codebase.
We sincerely thank the MoGE authors and contributors for their excellent work and open-source contribution.

📖 Citation

If you find this project useful in your research, please cite:

@inproceedings{lyu2026streamingdepth,
  title={Stabilizing Streaming Video Geometry via Dynamic Feature Normalization},
  author={Lyu, Xiaoyang and Liu, Muxin and Wu, Xiaoshan and Wang, Ruicheng and Huang, Yi-Hua and Sun, Yang-Tian and Shi, Shaoshuai and Qi, Xiaojuan},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year={2026}
}

Please also consider citing MoGE:

@inproceedings{wang2025moge,
  title={Moge: Unlocking accurate monocular geometry estimation for open-domain images with optimal training supervision},
  author={Wang, Ruicheng and Xu, Sicheng and Dai, Cassie and Xiang, Jianfeng and Deng, Yu and Tong, Xin and Yang, Jiaolong},
  booktitle={Proceedings of the Computer Vision and Pattern Recognition Conference},
  pages={5261--5271},
  year={2025}
}