anon
commited on
Update README.md
Browse files
README.md
CHANGED
|
@@ -96,6 +96,14 @@ Baseline PaddleOCR-VL performance on Polish test set:
|
|
| 96 |
| Exact Match | 74.00% |
|
| 97 |
| Diacritic Accuracy | 74.14% |
|
| 98 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 99 |
Key diacritic confusions in baseline:
|
| 100 |
- `ł` frequently confused with `l` or `t`
|
| 101 |
- `ę` sometimes rendered as `e`
|
|
@@ -118,7 +126,7 @@ If you use this model, please cite:
|
|
| 118 |
```bibtex
|
| 119 |
@misc{rysocr2024,
|
| 120 |
title={RysOCR: Polish OCR LoRA for PaddleOCR-VL},
|
| 121 |
-
author={
|
| 122 |
year={2024},
|
| 123 |
publisher={Hugging Face},
|
| 124 |
url={https://huggingface.co/anon13370/RysOCR}
|
|
|
|
| 96 |
| Exact Match | 74.00% |
|
| 97 |
| Diacritic Accuracy | 74.14% |
|
| 98 |
|
| 99 |
+
Improved version:
|
| 100 |
+
Summary:
|
| 101 |
+
| | Baseline | Fine-tuned |
|
| 102 |
+
|-------|----------|------------|
|
| 103 |
+
| CER | 5.58% | 1.60% |
|
| 104 |
+
| WER | 13.37% | 7.21% |
|
| 105 |
+
| Exact | 74% | 76% |
|
| 106 |
+
|
| 107 |
Key diacritic confusions in baseline:
|
| 108 |
- `ł` frequently confused with `l` or `t`
|
| 109 |
- `ę` sometimes rendered as `e`
|
|
|
|
| 126 |
```bibtex
|
| 127 |
@misc{rysocr2024,
|
| 128 |
title={RysOCR: Polish OCR LoRA for PaddleOCR-VL},
|
| 129 |
+
author={Kacper Wikieł},
|
| 130 |
year={2024},
|
| 131 |
publisher={Hugging Face},
|
| 132 |
url={https://huggingface.co/anon13370/RysOCR}
|