飞桨学习赛：文档图像摩尔纹消除

Who this competition fits

适合已有计算机视觉和图像复原基础、能用 PaddlePaddle 2.0 及以上版本训练端到端模型的个人参赛者。参赛者应能处理成对的源图像与 GT 图像，并生成与测试输入同名、同尺寸的输出图片；比赛仅支持个人参赛。

What this competition gives you

There is a baseline you can run as-is — starting from it is usually far faster than building from scratch.

• BaselineForkable PaddlePaddle competition baseline notebook
• DataCompetition training and test image datasets
• TutorialBaseline explanation video
• Reference paperMBCNN demoireing paper
• Reference paperWDNet demoireing paper
• Reference paperCamera-captured screen image demoireing paper
• Reference paperMSSSIM reference paper
• Reference paperAIM 2019 demoireing dataset study paper
• DataTIP 2018 dataset on Google Drive
• DataTIP 2018 dataset on Baidu Netdisk
• DataAIM 2019 validation moiré images
• DataAIM 2019 validation clear images
• DataAIM 2019 testing dataset
• BaselineMBCNN open-source reference code
• BaselineWDNet open-source reference code

From registration to a first submission

1. 01

   PaddlePaddle 2.0+

2. 02

   图像到图像模型

3. 03

   图像去摩尔纹

4. 04

   图像预处理与尺寸对齐

5. 05

   PSNR 与 MS-SSIM

6. 06

   ZIP 结果文件打包

Before you commit: 核心难点是在真实场景采集的屏摄图像中去除摩尔纹，同时尽量恢复原图细节，并兼顾未公开 GT 测试集上的泛化表现。排名同时考察像素误差和多尺度结构相似性，因此模型不能只做平滑降噪，还要保留视觉结构；提交还须严格满足同名同尺寸图片、根目录 ZIP 和 readme.txt 的格式要求。