COCO TS

COCO TS数据集是由意大利锡耶纳大学的研究团队创建，专门为场景文本分割任务提供像素级监督。该数据集包含14690张图像，这些图像来自COCO-Text数据集，通过弱监督学习方法生成像素级监督。创建过程中，利用背景-前景网络从合成数据中学习，进而为真实图像生成监督。COCO TS数据集主要应用于场景文本分割，旨在提高深度学习模型在真实图像上的性能，减少与合成数据训练的差距。

The COCO TS dataset was developed by a research team from the University of Siena in Italy, and is specifically constructed to provide pixel-level supervision for scene text segmentation tasks. It comprises 14,690 images sourced from the COCO-Text dataset, whose pixel-level supervision is generated using weakly-supervised learning approaches. During the dataset creation process, a background-foreground network was first trained on synthetic data, and subsequently used to generate supervision signals for real-world images. The COCO TS dataset is primarily applied to scene text segmentation tasks, with the objective of improving the performance of deep learning models on real-world images and narrowing the performance gap between models trained on synthetic datasets and those trained on real-world data.

搜集汇总

背景与挑战

背景概述

COCO TS数据集是一个专门用于场景文本分割任务的数据集，由意大利锡耶纳大学研究团队基于COCO-Text数据集创建，包含14690张图像，并通过弱监督学习方法生成像素级监督。它旨在提高深度学习模型在真实图像上的性能，减少与合成数据训练的差距。