Optical Character Recognition on Vertical Japanese prose
2.7Character Error RateEffOCR-T (Base)
Evaluation results
| Method | Links | |
|---|---|---|
| EffOCR-T (Base)Seq2Seq?=X, Transformer?=✓, Pretraining=from scratch, Parameters=101.8 M2023.04 | 2.7 | |
| EffOCR-C (Base)Seq2Seq?=X, Transformer?=X, Pretraining=from scratch, Parameters=112.5 M2023.04 | 3 | |
| EffOCR-C (Small)Seq2Seq?=X, Transformer?=X, Pretraining=from scratch, Parameters=9.3 M2023.04 | 3.6 | |
| Google Cloud Vision OCRPretraining=off-the-shelf2023.04 | 13.5 | |
| Baidu OCRPretraining=off-the-shelf2023.04 | 17.7 | |
| Tesseract OCR (Best)Pretraining=off-the-shelf, Parameters=1.4 M2023.04 | 74.4 |