EBMV @ ECCV 2026

Broad-Light Adaptive Brightness Adjustment

SEE Challenge 2026

Event-Guided Low-Level Imaging

RGB + events for brightness-adjusted RGB restoration under low light, over-exposure, mixed illumination, high contrast, and motion.

Competition Page Dataset Code Tutorial Submit Leaderboard

At a Glance

610K+

RGB images with event data

202

real-world scenarios

>1000x

illumination variation

3

low / normal / high light

4

lighting groups per scene on average

1.9M

SEE-Net baseline parameters

Challenge Visuals

SEE challenge task definition with challenging input, event guidance, and restored output
Problem Definition

Input RGB under challenging illumination + event representation -> brightness-adjusted RGB output.

SEE-600K dataset samples under low, normal, and high lighting conditions
Dataset Samples

Examples cover low-light, normal-light, high-light, mixed illumination, and event views.

SEE-600K scene examples and scenario word cloud
Scenario Coverage

202 real-world scenarios with broad scene categories and multiple lighting groups.

SEE baseline visualization results produced by the released code
Baseline Visualization

SEE-Net uses RGB frames, event data, and a brightness prompt for controllable output exposure.

Task & Evaluation

Task

Input Challenging RGB image, event stream / voxel, optional brightness prompt.
Output Brightness-adjusted RGB image matching the input sample ID and resolution.
Goal Well-exposed, detailed, structurally faithful, naturally colored result.

Metrics

Metric Role Better
PSNR Primary ranking Higher
SSIM Structure Higher
LPIPS Perception Lower

Getting Started

Code Guide

1. Data Download SEE-600K from Hugging Face. The Hugging Face release is aligned and ready to use.
2. Setup Clone SEE, create a Python 3.10 environment, then install requirements.txt.
3. Train Edit DATASET.root in options/SEE/SEENet_SEE.yaml, then run SEE-Net.
4. Test Use --TEST_ONLY=True and --VISUALIZE=True to save validation outputs.
Read Tutorial Open Code

Minimal Commands

git clone https://github.com/yunfanLu/SEE.git
cd SEE
conda create -n see python=3.10 -y
conda activate see
pip install -r requirements.txt

export PYTHONPATH="./":$PYTHONPATH
python see/main.py \
  --yaml_file="options/SEE/SEENet_SEE.yaml" \
  --log_dir="./logs/SEE/SEENet_SEE/" \
  --alsologtostderr=True

Submission

ZIP Structure

submission.zip
`-- results/
    |-- scene_000001.png
    |-- scene_000002.png
    |-- scene_000003.png
    `-- ...

Checklist

  • Filenames match input sample IDs.
  • Images are RGB PNG files.
  • Resolution matches input / reference resolution.
  • Folder structure remains unchanged.
  • No hidden test ground truth or manual test-set tuning.

Timeline

May 10, 2026

Challenge website opens

May 15, 2026

Validation server online

June 25, 2026

Test data and server online

July 3, 2026

Final submission deadline

July 10, 2026

Results announcement

Rules & Links

Rules

  • Use of SEE-600K is encouraged.
  • External data, pretrained models, and synthetic data are allowed if disclosed.
  • Hidden test ground truth must not be used.
  • Top teams may be asked for a technical report or code for verification.

Citation

SEE: See Everything Every Time - Adaptive Brightness Adjustment for Broad Light Range Images via Events

@article{lu2025SEE,
  title={SEE: See Everything Every Time - Adaptive Brightness Adjustment for Broad Light Range Images via Events},
  author={Yunfan Lu, Xiaogang Xu, Hao Lu, Yanlin Qian, Pengteng Li, Huizai Yao, Bin Yang, Junyi Li, Qianyi Cai, Weiyu Guo, Hui Xiong},
  year={2025},
}

Contact: Yunfan Lu

ylu066@connect.hkust-gz.edu.cn