Perception and Grounding

Restormer

Transformer-based high-resolution image restoration for denoising, deblurring, and deraining.

Tool Introduction

Core parameters, trigger timing, and visual before/after demo references.

Short Explanation

Restormer restores degraded images with a transformer architecture tuned for image quality and efficiency.

InputDegraded image

OutputRestored image

Trigger TimingTriggered on demand after the required input files and configuration are prepared.

RuntimePython / PyTorch

BeforeDegraded image

Prepare the scene, image, video, sensor stream, prompt, or configuration expected by the original project.

AfterRestored image

Read the produced visualization, prediction, map, trajectory, mask, grasp pose, or other documented artifact.

Preset Example

A quick-run style example for the documentation page.

Inputtools/restormer/examples/blur.png

Prompttask: Motion_Deblurring

ExpectedA restored image saved to the result directory.

Parameters And Output

Readable controls and the meaning of each returned artifact.

Parameter Explanation

taskselectMotion_Deblurring

Restoration task profile (deblurring, denoising, deraining, etc.).

input_dirpath

Input image directory.

result_dirpath

Directory where restored outputs are written.

Output Explanation

restored_image

Image after restoration processing.

How To Use

Official resources, deployment steps, academic context, citation, and source-reported benchmark numbers.

Resources

GitHubhttps://github.com/swz30/Restormer Code Downloadhttps://github.com/swz30/Restormer/archive/refs/heads/main.zip arXivhttps://arxiv.org/abs/2111.09881 Pretrained Modelshttps://github.com/swz30/Restormer#pretrained-models

Deployment Notes

Install dependencies and download task-specific pretrained weights.
Pick the matching task configuration for your degradation type.
Run demo/eval script with repository-relative directories.
Collect outputs under tools/restormer/runs/.

Relative Path Example

python demo.py --task Motion_Deblurring --input_dir tools/restormer/examples --result_dir tools/restormer/runs

Expected Result Shape

{
  "tool": "restormer",
  "status": "ok",
  "results": [
    {
      "label": "Image restoration",
      "score": 0.87,
      "output": "Restored image"
    }
  ],
  "timing": {
    "runtime": "Runtime depends on restoration task and image resolution; the paper tables emphasize PSNR/SSIM quality metrics.",
    "device": "documented in source benchmark when available"
  },
  "artifacts": {
    "visualization": "tools/restormer/runs/visualization.png",
    "raw_predictions": "tools/restormer/runs/predictions.json"
  }
}

Paper figure

Academic Info

Paper identity and contribution summary.

TitleRestormer: Efficient Transformer for High-Resolution Image Restoration

AuthorsSyed Waqas Zamir, Aditya Arora, Salman Khan, et al.

VenueCVPR 2022 / arXiv:2111.09881

ContributionDesigns an efficient transformer architecture specialized for high-resolution restoration tasks.

Citation

@misc{restormer2022,
  title={Restormer: Efficient Transformer for High-Resolution Image Restoration},
  author={Syed Waqas Zamir and Aditya Arora and Salman Khan and et al.},
  year={2022},
  note={CVPR 2022 / arXiv:2111.09881},
  url={https://arxiv.org/abs/2111.09881}
}

Benchmark

Only compact, source-reported numbers are shown here.

Dataset	Metric	Value	Runtime	Source
GoPro motion deblurring	PSNR / SSIM	32.92 / 0.961	Task-specific restoration model	Official CVPR 2022 paper
Image deraining average over Test100, Rain100H, Rain100L, Test2800, Test1200	Average PSNR / SSIM	33.96 / 0.935	Task-specific restoration model	Official CVPR 2022 paper, Table 1
Real image denoising, SIDD / DND	PSNR / SSIM	40.02 / 0.960 on SIDD; 40.03 / 0.956 on DND	Task-specific denoising model	Official CVPR 2022 paper, Table 6

Artifacts

Official CVPR 2022 paper tables, pretrained checkpoints, task demos, and evaluation scripts.

Demo Images

Visual references from the original tool. Click any image to inspect the original size.