飞桨学习赛：水印智能消除赛

Who this competition fits

适合已有计算机视觉和图像到图像训练基础、愿意使用 PaddlePaddle 2.0 及以上版本的个人参赛者。尤其适合想练习图像恢复/图像重建模型，以及模型部署与推理效率优化的人。

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 competition baseline notebook
• TutorialVideo tutorial with score-improvement strategies
• DataWatermark dataset with 1,012,550 training and 5,000 test samples
• BaselineNoise-to-noise watermark removal source code reference
• BaselineSLBR visible watermark removal source code reference
• BaselineAutomatic watermark detection source code reference
• BaselineWatermark data generation source code reference
• Reference paperMS-SSIM reference paper

From registration to a first submission

1. 01

   PaddlePaddle 2.0+

2. 02

   图像到图像模型

3. 03

   图像恢复

4. 04

   JPEG图像读写

5. 05

   模型推理脚本编写

6. 06

   ZIP提交打包

Before you commit: 难点不只是去除水印并恢复背景细节，还要兼顾未公开测试集上的泛化效果和实际推理性能。提交的是可自动运行的模型与 predict.py，输出图片必须保持原文件名和尺寸；单张推理耗时达到 1.2 秒时，决赛性能分数为 0。