COCONut Dataset

Introducing COCONut, a cutting-edge, comprehensive segmentation dataset boasting approximately 383K images and 5.18M meticulously verified panoptic segmentation masks. Complementing this expansive resource is COCONut-val, a novel validation set designed to serve as a challenging testbed for the community.

This dataset caters to a wide array of tasks, including panoptic, instance, and semantic segmentation, object detection, pretraining for open-vocabulary segmentation, and mask-conditional image synthesis. As we delve into the Era of Big Data, we anticipate the emergence of even more innovative tasks facilitated by the vast scale and versatility of the COCONut dataset. We invite the community to explore its potential and contribute to advancements in the field.


- The Kaggle website only hosts our meticulously crafted annotations. The users are required to download the COCO images from the official COCO website. Moreover, to support forthcoming research endeavors, we have furnished benchmarked results on COCONut, accompanied by comprehensive tutorials available on our GitHub repository (https://github.com/bytedance/coconut_cvpr2024). For further details, we encourage the users to consult the resources provided therein.

隆重推出 COCONut 数据集，该数据集堪称前沿之作，内容全面，包含约 383K 张图像及 5.18M 张经过细致验证的全面分割掩码。与之相辅相成的 COCONut-val 验证集，旨在为社区提供一个具有挑战性的测试平台。本数据集适用于众多任务，包括全面分割、实例分割、语义分割、目标检测、开放词汇分割的预训练以及掩码条件图像合成。随着我们步入大数据时代，我们预期 COCONut 数据集的广泛规模和多样性将催生出更多创新任务。我们诚挚邀请社区共同探索其潜力，为该领域的发展贡献力量。


- Kaggle 网站仅提供我们精心制作的标注。用户需从官方 COCO 网站下载 COCO 图像。此外，为支持未来的研究工作，我们已在 COCONut 上提供基准测试结果，并附上详尽的教程，可在我们的 GitHub 仓库（https://github.com/bytedance/coconut_cvpr2024）中找到。更多详细信息，请参阅相关资源。

搜集汇总

背景与挑战

背景概述

COCONut是一个包含约383K图像和5.18M panoptic分割掩码的大规模图像分割数据集，附带一个专门设计的验证集COCONut-val。该数据集支持多种计算机视觉任务，包括分割、检测和图像合成等，适用于大数据时代的研究和创新。