Dharma-AI/DharmaOCR-Benchmark
收藏资源简介:
DharmaOCR-Benchmark是一个包含496个实例的评估套件,专注于巴西葡萄牙语文档的OCR模型。它涵盖了印刷文本、手写文本和法律/行政文档,这些领域在现有基准测试(如OCRBench和olmOCR-Bench)中代表性不足。该基准测试不仅评估转录质量,还将文本退化率和单位推理成本作为首要指标。数据集分为三个子集:ESTER-Pt(363个样本,巴西葡萄牙语印刷文本识别)、Legal(83个样本,法律和行政文档)和BRESSAY(50个样本,巴西葡萄牙语手写文本识别)。评估协议包括基于LevenshteinRatio和BLEU的复合分数,以及文本退化率和每页单位成本等附加指标。
DharmaOCR-Benchmark is a 496-instance evaluation suite for OCR models focused on Brazilian Portuguese documents. It covers printed text, handwritten text, and legal/administrative documents — domains underrepresented in existing benchmarks like OCRBench and olmOCR-Bench. This benchmark evaluates not only transcription quality, but also text degeneration rate and unit inference cost as first-class metrics. The dataset is composed of three subsets: ESTER-Pt (363 samples, printed text recognition in Brazilian Portuguese), Legal (83 samples, legal and administrative documents), and BRESSAY (50 samples, handwritten text recognition in Brazilian Portuguese). The evaluation protocol includes a composite score based on LevenshteinRatio and BLEU, along with additional metrics like text degeneration rate and unit cost per page.





