Perception and Grounding

Zero-DCE

Zero-reference deep curve estimation for enhancing low-light images.

Tool Introduction

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

Short Explanation

Zero-DCE brightens and enhances low-light images using a zero-reference training objective.

InputLow-light image

OutputEnhanced image

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

RuntimePython / PyTorch

BeforeLow-light image

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

AfterEnhanced 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/zero-dce/examples/lowlight.png

Promptdefault enhancement pipeline

ExpectedEnhanced image with improved brightness and contrast.

Parameters And Output

Readable controls and the meaning of each returned artifact.

Parameter Explanation

inputpath

Input low-light image path or directory.

outputpath

Directory for enhanced outputs.

Output Explanation

enhanced_image

Brightness- and contrast-improved output image.

How To Use

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

Resources

GitHubhttps://github.com/Li-Chongyi/Zero-DCE Code Downloadhttps://github.com/Li-Chongyi/Zero-DCE/archive/refs/heads/master.zip arXivhttps://arxiv.org/abs/2001.06826 Pretrained Modelhttps://github.com/Li-Chongyi/Zero-DCE#testing

Deployment Notes

Install project dependencies and download official pretrained weights.
Prepare low-light image directory under tools/zero-dce/examples/.
Run test script with repository-relative input/output paths.
Store enhanced results under tools/zero-dce/runs/.

Relative Path Example

python lowlight_test.py --input tools/zero-dce/examples --output tools/zero-dce/runs

Expected Result Shape

{
  "tool": "zero-dce",
  "status": "ok",
  "results": [
    {
      "label": "Low-light image enhancement",
      "score": 0.87,
      "output": "Enhanced image"
    }
  ],
  "timing": {
    "runtime": "The official paper reports about 500 FPS for 640x480x3 images on GPU, 0.0025 s for 1200x900x3 in Table 3, and 30 minutes training.",
    "device": "documented in source benchmark when available"
  },
  "artifacts": {
    "visualization": "tools/zero-dce/runs/visualization.png",
    "raw_predictions": "tools/zero-dce/runs/predictions.json"
  }
}

Paper figure

Academic Info

Paper identity and contribution summary.

TitleZero-Reference Deep Curve Estimation for Low-Light Image Enhancement

AuthorsChongyi Li, Chunle Guo, Chen Change Loy

VenueCVPR 2020 / arXiv:2001.06826

ContributionEnhances low-light images without paired supervision via learnable curve estimation.

Citation

@misc{zerodce2020,
  title={Zero-Reference Deep Curve Estimation for Low-Light Image Enhancement},
  author={Chongyi Li and Chunle Guo and Chen Change Loy},
  year={2020},
  note={CVPR 2020 / arXiv:2001.06826},
  url={https://arxiv.org/abs/2001.06826}
}

Benchmark

Only compact, source-reported numbers are shown here.

Dataset	Metric	Value	Runtime	Source
Full-reference low-light enhancement comparison	PSNR / SSIM / MAE	16.57 / 0.59 / 98.78	PyTorch GPU	Official CVPR 2020 paper, Table 2
Runtime comparison, 1200x900x3 image	Runtime	0.0025 s	PyTorch GPU	Official CVPR 2020 paper, Table 3
Real-time enhancement claim	Throughput / training time	about 500 FPS for 640x480x3 images on GPU; 30 minutes training	GPU	Official CVPR 2020 paper
No-reference image sets	User study / perceptual index average	3.70 / 2.88	Quality evaluation	Official CVPR 2020 paper, Table 1

Artifacts

Official CVPR 2020 paper, runtime table, quality tables, official code, pretrained weights, and test scripts.

Demo Images

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