W-Bench|水印技术数据集|图像编辑数据集
收藏Robust Watermarking Using Generative Priors Against Image Editing: From Benchmarking to Advances
数据集概述
- 标题: Robust Watermarking Using Generative Priors Against Image Editing: From Benchmarking to Advances
- 作者: Shilin Lu, Zihan Zhou, Jiayou Lu, Yuanzhi Zhu, Adams Wai-Kin Kong
- 摘要: 当前的图像水印方法在面对大规模文本到图像模型启用的先进图像编辑技术时显得脆弱。这些模型在编辑过程中可能会扭曲嵌入的水印,对版权保护构成重大挑战。本文介绍了W-Bench,这是第一个全面评估水印方法对各种图像编辑技术(包括图像再生、全局编辑、局部编辑和图像到视频生成)的鲁棒性的基准。通过对十一种代表性水印方法在常见编辑技术下的广泛评估,我们发现大多数方法在编辑后无法检测到水印。为了解决这一局限性,我们提出了VINE,一种显著增强对各种图像编辑技术鲁棒性的水印方法,同时保持高图像质量。我们的方法涉及两个关键创新:(1)我们分析了图像编辑的频率特性,并发现模糊失真具有相似的频率特性,这使我们能够在训练期间使用它们作为代理攻击来增强水印鲁棒性;(2)我们利用大规模预训练的扩散模型SDXL-Turbo,将其适应于水印任务,以实现更不可察觉和鲁棒的水印嵌入。实验结果表明,我们的方法在各种图像编辑技术下实现了出色的水印性能,在图像质量和鲁棒性方面均优于现有方法。
内容
- 环境设置:
- 创建Conda环境
- 下载VINE检查点
- 推理:
- 水印编码
- 图像编辑
- 水印解码
- 质量指标计算
- 演示
- W-Bench: 即将推出
- 引用: 如果发现该仓库有用,请考虑引用
环境设置
创建Conda环境
git clone https://github.com/Shilin-LU/VINE.git conda env create -f environment.yaml conda activate vine cd diffusers pip install -e .
下载VINE检查点
模型VINE-B和VINE-R已发布,可在此处下载,并放置在./ckpt
文件夹中。
推理
水印编码
使用以下命令将消息编码到图像中:
python src/watermark_encoding.py --ckpt_path ./ckpt/VINE-R
--input_path ./example/input/2.png
--output_dir ./example/watermarked_img
--message Hello World!
图像编辑
提供UltraEdit和图像反演进行图像编辑,更多选项即将添加。使用以下命令编辑图像:
python src/image_editing.py --model ultraedit
--input_path ./example/watermarked_img/2_wm.png
--output_dir ./example/edited_watermarked_img
水印解码
使用以下命令从已编辑的水印图像中解码消息:
python src/watermark_decoding.py --ckpt_path ./ckpt/VINE-R
--input_path ./example/edited_watermarked_img/2_wm_edit.png
--groundtruth_message Hello World!
质量指标计算
使用以下命令计算单张图像的质量指标(PSNR、SSIM和LPIPS):
python src/quality_metrics.py --input_path ./example/input/2.png --wmed_input_path ./example/watermarked_img/2_wm.png
演示
提供一个完整的演示,包括水印编码、图像编辑、水印解码和质量指标计算的过程,请参考./src/demo.ipynb
获取详细说明。
W-Bench
即将推出
引用
如果发现该仓库有用,请考虑引用。

- 1Robust Watermarking Using Generative Priors Against Image Editing: From Benchmarking to Advances南洋理工大学 · 2024年
Population and Housing Census of 2007 - Ethiopia
Geographic coverage --------------------------- National coverage Analysis unit --------------------------- Household Person Housing unit Universe --------------------------- The census has counted people on dejure and defacto basis. The dejure population comprises all the persons who belong to a given area at a given time by virtue of usual residence, while under defacto approach people were counted as the residents of the place where they found. In the census, a person is said to be a usual resident of a household (and hence an area) if he/she has been residing in the household continuously for at least six months before the census day or intends to reside in the household for six months or longer. Thus, visitors are not included with the usual (dejure) population. Homeless persons were enumerated in the place where they spent the night on the enumeration day. The 2007 census counted foreign nationals who were residing in the city administration. On the other hand all Ethiopians living abroad were not counted. Kind of data --------------------------- Census/enumeration data [cen] Mode of data collection --------------------------- Face-to-face [f2f] Research instrument --------------------------- Two type sof questionnaires were used to collect census data: i) Short questionnaire ii) Long questionnaire Unlike the previous censuses, the contents of the short and long questionnaires were similar both for the urban and rural areas as well as for the entire city. But the short and the long questionnaires differ by the number of variables they contained. That is, the short questionnaire was used to collect basic data on population characteristics, such as population size, sex, age, language, ethnic group, religion, orphanhood and disability. Whereas the long questionnaire includes information on marital status, education, economic activity, migration, fertility, mortality, as well as housing stocks and conditions in addition to those questions contained in a short questionnaire.
catalog.ihsn.org 收录
China Health and Nutrition Survey (CHNS)
China Health and Nutrition Survey(CHNS)是一项由美国北卡罗来纳大学人口中心与中国疾病预防控制中心营养与健康所合作开展的长期开放性队列研究项目,旨在评估国家和地方政府的健康、营养与家庭计划政策对人群健康和营养状况的影响,以及社会经济转型对居民健康行为和健康结果的作用。该调查覆盖中国15个省份和直辖市的约7200户家庭、超过30000名个体,采用多阶段随机抽样方法,收集了家庭、个体以及社区层面的详细数据,包括饮食、健康、经济和社会因素等信息。自2011年起,CHNS不断扩展,新增多个城市和省份,并持续完善纵向数据链接,为研究中国社会经济变化与健康营养的动态关系提供了重要的数据支持。
www.cpc.unc.edu 收录
MedDialog
MedDialog数据集(中文)包含了医生和患者之间的对话(中文)。它有110万个对话和400万个话语。数据还在不断增长,会有更多的对话加入。原始对话来自好大夫网。
github 收录
BBGRE
The Brain & Body Genetic Resource Exchange (BBGRE) provides a resource for investigating the genetic basis of neurodisability. It combines phenotype information from patients with neurodevelopmental and behavioural problems with clinical genetic data, and displays this information on the human genome map.
国家生物信息中心 收录
URPC系列数据集, S-URPC2019, UDD
URPC系列数据集包括URPC2017至URPC2020DL,主要用于水下目标的检测和分类。S-URPC2019专注于水下环境的特定检测任务。UDD数据集信息未在README中详细描述。
github 收录