CA-sup
收藏资源简介:
本研究提出了一种名为CA-sup的数据集,由四川大学和Megvii Technology联合创建。该数据集包含80万对图像,用于监督式单应性学习。数据集通过迭代框架生成,结合了内容一致性模块和质量评估模块,确保数据质量。该数据集旨在解决传统监督学习方法中训练数据不足的问题,通过提供高质量的训练样本,提升单应性估计网络的性能,使其更好地适应真实世界场景。
This study presents a dataset named CA-sup, jointly developed by Sichuan University and Megvii Technology. This dataset consists of 800,000 image pairs, tailored for supervised homography learning. It is generated through an iterative framework that incorporates a content consistency module and a quality assessment module to guarantee high data quality. The proposed dataset aims to mitigate the problem of insufficient training data in conventional supervised learning approaches. By supplying high-quality training samples, it improves the performance of homography estimation networks, allowing them to better adapt to real-world scenarios.

- 1Supervised Homography Learning with Realistic Dataset Generation四川大学 · 2023年



