飞桨学习赛：手写文字擦除

Who this competition fits

适合已能用 PaddlePaddle 训练端到端视觉模型的开发者，尤其是想练习图像到图像转换、图像修复或场景文字擦除的人。个人即可参赛，且页面说明面向全社会开放；但提交模型必须使用 PaddlePaddle 2.0 及以上版本。

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 baseline notebook
• DataCompetition training and test image datasets
• Reference paperMSSSIM reference paper
• Reference paperEraseNet reference paper
• Reference paperEnsNet reference paper
• Reference paperCascaded text removal reference paper
• Reference paperMTRNet++ reference paper
• Reference paperMTRNet reference paper
• DataSCUT-EnsText public dataset
• DataSynthText public dataset
• BaselineEraseNet open-source code
• BaselineOfficial SceneTextRemoval TensorFlow implementation
• BaselineMinimal SceneTextRemover PyTorch implementation
• BaselineMTRNet/MTRNet++ open-source code
• BaselineEnsNet open-source code

From registration to a first submission

1. 01

   PaddlePaddle 2.0+

2. 02

   图像到图像模型

3. 03

   图像修复与文本擦除

4. 04

   PSNR 与 MS-SSIM 优化

5. 05

   PNG 图像批量处理

6. 06

   ZIP 提交文件组织

Before you commit: 难点是既要去除试卷图像中的手写痕迹，又要精确恢复被笔迹遮挡的原始内容；测试集 GT 不公开，需让模型在真实场景处理而来的图像上保持泛化。除训练效果外，还必须严格生成与输入同名、同尺寸的 PNG，并将结果与 readme.txt 以无文件夹层级的 ZIP 提交。