DIV2K
收藏资源简介:
DIV2K数据集分为: 列车数据: 从800高清高分辨率图像开始,我们获得相应的低分辨率图像,并为2、3和4个降尺度因子提供高分辨率和低分辨率图像 验证数据: 100高清晰度高分辨率图像用于生成低分辨率对应图像,低分辨率从挑战开始提供,并用于参与者从验证服务器获得在线反馈; 当挑战的最后阶段开始时,高分辨率图像将被释放。 测试数据: 100多样的图像用于生成低分辨率的相应图像; 参与者将在最终评估阶段开始时收到低分辨率图像,并在挑战结束并确定获胜者后宣布结果。
The DIV2K dataset is divided into three parts: Training data: Starting with 800 high-definition high-resolution images, we generate their corresponding low-resolution images, and provide pairs of high-resolution and low-resolution images for downscaling factors of 2, 3, and 4. Validation data: 100 high-definition high-resolution images are used to generate their corresponding low-resolution counterparts. The low-resolution images are provided at the start of the challenge and are used by participants to obtain online feedback via the validation server; the high-resolution images will be released when the final phase of the challenge begins. Test data: 100 diverse images are used to generate their corresponding low-resolution versions. Participants will receive the low-resolution images at the start of the final evaluation phase, and the results will be announced after the challenge concludes and the winner is determined.

- DIV2K数据集首次发表,作为高分辨率图像超分辨率任务的标准基准数据集。
- DIV2K数据集在多个国际计算机视觉会议上被广泛引用和讨论,成为图像超分辨率研究的重要参考。
- DIV2K数据集的应用扩展到视频超分辨率领域,推动了相关技术的进步。
- DIV2K数据集的改进版本发布,增加了更多的图像样本和多样性,进一步提升了其在研究中的价值。
- DIV2K数据集在深度学习框架中的集成,使得研究人员能够更方便地进行实验和比较。
- 1DIV2K: A Dataset for Image Super-ResolutionETH Zurich · 2017年
- 2Deep Learning for Image Super-Resolution: A SurveyUniversity of Science and Technology of China · 2019年
- 3ESRGAN: Enhanced Super-Resolution Generative Adversarial NetworksTsinghua University · 2018年
- 4Image Super-Resolution Using Deep Convolutional NetworksUniversity of Southern California · 2016年
- 5Real-World Super-Resolution via Kernel Estimation and Noise InjectionTsinghua University · 2020年



