Gyro-based Deep Video Deblurring

Project | Paper | Supple

Official Implementation of CVPR 2026 Paper

Gyro-based Deep Video Deblurring
Jaesung Rim¹, Woohyeok Kim¹, Haeyun Lee², Heemin Yang¹, Ke Wang³, Sunghyun Cho^1
1POSTECH, ²KOREATECH, ³Pika Labs
IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2026

Install

conda create -n GyroDVD python=3.8
conda activate GyroDVD
conda install pytorch==1.13.1 torchvision==0.14.1 torchaudio==0.13.1 pytorch-cuda=11.7 -c pytorch -c nvidia
pip install -r requirements.txt
pip install git+https://github.com/cheind/pytorch-debayer
python setup.py develop

Download

Dataset [Hugging Face]

Descriptions (click)

For the training set, noise and saturation pixels are synthesized on-the-fly in GyroDVD_model.py. Please refer to RSBlurPipeline.

For the validation and test sets, noise and saturation pixels have already been applied and are included in the released dataset.

• GyroVD_Syn_train.tar.gz : 50,500 frames for training.
  • GyroVD_Syn_train_tau.tar.gz : pre-computed tau. (only required for training.)
• GyroVD_Syn_val.tar.gz : 5,000 frames for validation.
  • GyroVD_Syn_val_tau.tar.gz : pre-computed tau for validation set. (only required for training.)
• GyroVD_Syn_test.tar.gz : 7,700 frames for evaluation.
  • GyroVD_Syn_test_tau.tar.gz : pre-computed tau for test set. (only required for training.)
• GyroVD_Real.tar.gz: 10,000 real-world frames for evaluation.

To facilitate future research, we also provide accelerometer and magnetometer data.

The GyroVD-Syn dataset

# GyroVD_Syn_train.tar.gz
GyroVD_Syn_train
├── 0413/VID_20250412_101526
│   ├── blur # blurred frames
│   │   ├── 000001_160466752524897_160466785733026_160466786578871_avg09_blur.png
│   │   ├── 000001_(exp_start_1th)_(exp_start_9th)_(exp_end)_avg09_blur.png
│   │   ...
│   ├── gt # gt frames
│   │   ├── 000001_160466752524897_160466785733026_160466786578871_avg09_gt.png
│   │   ...
│   ├── meta_info
│   │   ├── gyro.csv # gyroscope data
│   │   ├── accel.csv # accelerometer data
│   │   ├── magnetic.csv # magnetometer data
│   │   ├── cam_metainfo.csv # camera meta data
│   ├── rot # pre-computed rotation matrices
│   ├── sat_mask # saturation mask for RSBlur pipeline
├── 0413/VID_20250412_101616
│   ...
...

The GyroVD-Real dataset

# GyroVD_Real.tar.gz
GyroVD-Real
├── 1029_GP/VID_20251029_013931
│   ├── blur # blurred frames
│   │   ├── 000001_193795012257259_193795012257259_193795111113483_blur.jpg
│   │   ...
│   ├── meta_info
│   │   ├── gyro.csv # gyroscope data
│   │   ├── accel.csv # accelerometer data
│   │   ├── magnetic.csv # magnetometer data
│   │   ├── cam_metainfo.csv # camera meta data
│   ...
...

Deblurred Results [Hugging Face]

Pre-trained models [link]

Descriptions (click)

• GyroDVD_48.pth: Weight of GyroDVD-48.
• GyroDVD_64.pth: Weight of GyroDVD-64.
• GyroDVD_96.pth: Weight of GyroDVD-96.
• GyroDVD_128.pth: Weight of GyroDVD-128.
• raft-small.pth: Weight of RAFT_small for optical flow estimation.

Demo

# demo of samples from GyroVD-Real and GyroVD-Syn
python inference_GyroVDReal.py --dataset_root=demo/GyroVD_Real  --out_path=results/GyroDVD_128_Real_demo --model_size=128
python inference_GyroVDSyn.py --dataset_root=demo/GyroVD_Syn  --out_path=results/GyroDVD_128_Syn_demo --model_size=128

Testing

# datasets should be located in dataset
# pre-trained weights should be located in model_zoos

## test on GyroVD-Real
python inference_GyroVDReal.py --dataset_root=dataset/GyroVD_Real  --out_path=results/GyroDVD_128_Real --model_size=128

## test on GyroVD-Syn
python inference_GyroVDSyn.py --dataset_root=dataset/GyroVD_Syn_test  --out_path=results/GyroDVD_128_Syn --model_size=128

Evaluation

# ./GyroDVD

# compute PSNR and SSIM on GyroVD-Syn
python evaluation/eval_Syn/evaluate_GyroVD_Syn.py --input_dir=results/GyroDVD_128_Syn --gt_root=dataset/GyroVD_Syn_test --out_txt=results/GyroDVD_128_Syn.txt

# Compute TOP-IQA on GyroVD-Real. Requires IQA-PyTorch.
python evaluation/eval_Real/inference_iqa_resize.py -t=results/GyroDVD_128_Real -m=topiq_nr --save_file=results/GyroDVD_128_Real_topiq.txt;

# Compute BRISQUE and NIQE on GyroVD-Real. Tested on MATLAB R2023
# We found that BRISQUE and NIQE in IQA-PyTorch are sometimes not robust, so we use the MATLAB implementations for evaluation.
# ./evaluation/eval_Real
addpath('compute_iqa_matlab'); 
compute_iqa_matlab('../../results/GyroDVD_128_Real', 'brisque', '../../results/GyroDVD_128_Real_brisque.txt');
compute_iqa_matlab('../../results/GyroDVD_128_Real', 'niqe', '../../results/GyroDVD_128_Real_niqe.txt');

Training

# Download the GyroVD_Syn_train_tau dataset before training.

# GyroDVD is trained with batch size of 4
CUDA_VISIBLE_DEVICES=0,1,2,3 python -m torch.distributed.launch --nproc_per_node=4 --master_port=4016 basicsr/train.py -opt options/train/GyroDVD_128_train.yml --auto_resume --launcher pytorch

License

The GyroVD dataset is released under CC BY 4.0 license.

Acknowledgements

The code is based on BasicSR, RAFT, ShiftNet and IQA-PyTorch.

Citation

If you use our dataset and code for your research, please cite our paper.

@inproceedings{GyroDVD_rim,
 title={Gyro-based Deep Video Deblurring},
 author={Jaesung Rim and Woohyeok Kim and Haeyun Lee and Heemin Yang and Ke Wang and Sunghyun Cho},
 booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
 year={2026}
}