Subset of AdobeVFR. The dataset contains "real-world text images".

We collected 201,780 text images from various typography forums, where people post these images seeking help from experts to identify the fonts. Most of them come with hand-annotated font labels which may be inaccurate. (...) Finally, we obtain 4,384 real-world test images with reliable labels, covering 617 classes (out of 2,383). (...) Removing the 4,384 labeled images from the full set, we are left with 197,396 unlabeled real- world images which we denote as VFR_real_u.

The labeled images form VFR_real_test. The other part of AdobeVFR consists of synthetic data (with 2383 classes).