VFR-Wild Dataset

325 word images intended for font recognition, whose fonts are included in VFR-447 (and VFR-2420).

(...) 325 real world test images for the font classes we have in the training set. These images were collected from typography forums, such as myfonts.com, where people post these images seeking help from experts to identify the fonts. Compared with the synthetic data, these images typically have much larger appearance variations caused by scale, background, lighting, noise, perspective distortions, and compression artifacts. We manually cropped the texts from these images with a bounding box to normalize the text size approximately to the same scale as the synthetic data.

本数据集包含325张旨在用于字体识别的图像，其中所涉及的字体均收录于VFR-447（及VFR-2420）字体库中。 (... ) 此外，该数据集还提供了针对训练集中字体类别的325张现实世界测试图像。这些图像源自于诸如myfonts.com等字体论坛，用户在此类论坛上发布图像，寻求专家的帮助以识别字体。与合成数据相比，这些图像通常具有更大的外观变化，这是由尺度、背景、光照、噪声、透视畸变和压缩伪影等因素引起的。我们通过手动裁剪图像中的文本，并使用边界框进行规范化处理，使文本尺寸大致与合成数据保持相同的比例。