Parameter-Efficient Fine-Tuning of DINOv2 for Large-Scale Font Classification
This work addresses a gap in font classification benchmarks for open-source typefaces, providing a reproducible resource for the typography and computer vision communities, though it is incremental in applying existing methods to a new dataset.
The authors tackled the problem of font classification by creating GoogleFontsBench, the first public benchmark for open-source web fonts, and achieved 99.0% top-1 accuracy using LoRA fine-tuning on DINOv2 with only 1% of parameters trained.
We introduce GoogleFontsBench, the first public benchmark for classifying open-source web fonts, addressing a gap left by existing benchmarks that cover only commercial typefaces. GoogleFontsBench comprises 394 font variants across 32 Google Fonts families, a reproducible synthetic data generation pipeline (~575 images per variant, ~226K total), and a typographically-grounded evaluation metric (SWER) that weights errors by visual severity. We establish baselines using six fine-tuning strategies on a DINOv2 Vision Transformer backbone. Parameter-efficient adaptation with LoRA achieves 99.0% top-1 accuracy while training only 1% of the model's 87.2M parameters, with errors 140x less severe than random guessing. We release the benchmark, all trained models, and the full training pipeline as open-source resources.