Syn4Removal
收藏资源简介:
Syn4Removal是由中国科学技术大学和微软亚洲研究院联合创建的大规模对象移除数据集,旨在支持基于Masked-Region Guidance范式的对象移除任务。该数据集包含100万组图像三元组,每组包括原始背景图像、粘贴对象的掩码以及移除对象后的真实背景图像。数据集的创建过程涉及从公开的实例分割数据集中提取对象实例,并将其粘贴到不同的背景图像上,确保对象与背景的合理融合。通过这种方式,Syn4Removal提供了多样化的场景和对象类型,支持模型在复杂场景中准确移除对象并保持周围环境的连贯性。该数据集的应用领域主要集中在图像编辑和对象移除任务,旨在解决现有方法在移除对象时可能导致的背景失真或对象再生问题。
Syn4Removal is a large-scale object removal dataset jointly created by the University of Science and Technology of China and Microsoft Research Asia, designed to support object removal tasks based on the Masked-Region Guidance paradigm. The dataset contains one million image triplets, each including the original background image, a mask of the pasted object, and the true background image after the object has been removed. The creation process of the dataset involves extracting object instances from publicly available instance segmentation datasets, pasting them onto different background images, and ensuring a reasonable integration of the object with the background. Through this method, Syn4Removal provides diverse scenarios and object types, supporting models in accurately removing objects in complex scenes while maintaining the coherence of the surrounding environment. The application domain of this dataset is primarily focused on image editing and object removal tasks, aiming to address the potential issues of background distortion or object regeneration that may arise during object removal with existing methods.

- 1SmartEraser: Remove Anything from Images using Masked-Region Guidance中国科学技术大学, 微软亚洲研究院 · 2025年



