edit: model card
Browse files
README.md
CHANGED
@@ -3,4 +3,46 @@ language:
|
|
3 |
- ko
|
4 |
tags:
|
5 |
- ocr
|
6 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
- ko
|
4 |
tags:
|
5 |
- ocr
|
6 |
+
widget:
|
7 |
+
- src: https://raw.githubusercontent.com/ddobokki/ocr_img_example/master/g.jpg
|
8 |
+
example_title: word1
|
9 |
+
- src: https://raw.githubusercontent.com/ddobokki/ocr_img_example/master/khs.jpg
|
10 |
+
example_title: word2
|
11 |
+
- src: https://raw.githubusercontent.com/ddobokki/ocr_img_example/master/m.jpg
|
12 |
+
example_title: word3
|
13 |
+
---
|
14 |
+
|
15 |
+
# korean trocr model
|
16 |
+
|
17 |
+
## train datasets
|
18 |
+
AI Hub
|
19 |
+
- [๋ค์ํ ํํ์ ํ๊ธ ๋ฌธ์ OCR](https://aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&aihubDataSe=realm&dataSetSn=91)
|
20 |
+
- [๊ณต๊ณตํ์ ๋ฌธ์ OCR](https://aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&aihubDataSe=realm&dataSetSn=88)
|
21 |
+
|
22 |
+
## model structure
|
23 |
+
- encoder : [trocr-base-stage1's encoder](https://huggingface.co/microsoft/trocr-base-stage1)
|
24 |
+
- decoder : [KR-BERT-char16424](https://huggingface.co/snunlp/KR-BERT-char16424)
|
25 |
+
|
26 |
+
## how to use
|
27 |
+
|
28 |
+
```python
|
29 |
+
from transformers import TrOCRProcessor, VisionEncoderDecoderModel, AutoTokenizer
|
30 |
+
import requests
|
31 |
+
import unicodedata
|
32 |
+
from io import BytesIO
|
33 |
+
from PIL import Image
|
34 |
+
|
35 |
+
processor = TrOCRProcessor.from_pretrained("ddobokki/ko-trocr")
|
36 |
+
model = VisionEncoderDecoderModel.from_pretrained("ddobokki/ko-trocr")
|
37 |
+
tokenizer = AutoTokenizer.from_pretrained("ddobokki/ko-trocr")
|
38 |
+
|
39 |
+
url = "https://raw.githubusercontent.com/ddobokki/ocr_img_example/master/g.jpg"
|
40 |
+
response = requests.get(url)
|
41 |
+
img = Image.open(BytesIO(response.content))
|
42 |
+
|
43 |
+
pixel_values = processor(img, return_tensors="pt").pixel_values
|
44 |
+
generated_ids = model.generate(pixel_values, max_length=64)
|
45 |
+
generated_text = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
|
46 |
+
generated_text = unicodedata.normalize("NFC", generated_text)
|
47 |
+
print(generated_text)
|
48 |
+
```
|