New Font Recognition Benchmark tests VLMs on font ID with an easy set of ten sentences and a hard Stroop‑style set showing fifteen font names. Models performed near chance on the hard version. https://getnews.me/vision-language-models-struggle-with-font-recognition-study-finds/ #fontrecognition #vlm

We've expanded the AdobeVFR dataset and trained a FasterViT-2 model on millions of font samples — achieving state-of-the-art font recognition performance, that also powers smart downstream tasks.

Check out the results: col.la/fontrecognit...

#FontRecognition #ICME #AI #OpenSource