Seeing the Unseen: A Frequency Prompt Guided Transformer for Image Restoration (ECCV 2024)

Shihao Zhou, Jinshan Pan, Jinglei Shi, Duosheng Chen, Lishen Qu and Jufeng Yang

News

• Jul 02, 2024: FPro has been accepted to ECCV 2024 🎉

---

Training

Derain

To train FPro on SPAD, you can run:

./train.sh Deraining/Options/Deraining_FPro_spad.yml

Dehaze

To train FPro on SOTS, you can run:

./train.sh Dehaze/Options/RealDehazing_FPro.yml

Deblur

To train FPro on GoPro, you can run:

./train.sh Motion_Deblurring/Options/Deblurring_FPro.yml

Deraindrop

To train FPro on AGAN, you can run:

./train.sh Deraining/Options/RealDeraindrop_FPro.yml

Demoire

To train FPro on TIP18, you can run:

./train.sh Demoiring/Options/RealDemoiring_FPro.yml

Evaluation

To evaluate FPro, you can refer commands in 'test.sh'

For evaluate on each dataset, you should uncomment corresponding line.

Results

Experiments are performed for different image processing tasks including, rain streak removal, raindrop removal, haze removal, motion blur removal, and moire pattern removal. Here is a summary table containing hyperlinks for easy navigation:

Benchmark	Pretrained model	Visual Results
SPAD	(code:gd8j)	(code:ntgp)
AGAN	(code:dqml)	(code:ul55)
SOTS	(code:aagq)	(code:9ssj)
GoPro	(code:lhds)	(code:764e)
TIP18	(code:l13v)	(code:9und)

Citation

If you find this project useful, please consider citing:

@inproceedings{zhou_ECCV2024_FPro,
  title={Seeing the Unseen: A Frequency Prompt Guided Transformer for Image Restoration},
  author={Zhou, Shihao and Pan, Jinshan and Shi, Jinglei and Chen, Duosheng and Qu, Lishen and Yang, Jufeng},
  booktitle={ECCV},
  year={2024}
}

Acknowledgement

This code borrows heavily from Restormer.