podongchip commited on
Commit
ed139ac
ยท
verified ยท
1 Parent(s): ada9a77

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +69 -0
README.md ADDED
@@ -0,0 +1,69 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ base_model: juliozhao/DocLayout-YOLO-DocStructBench
4
+ tags:
5
+ - object-detection
6
+ - document-layout-analysis
7
+ - onnx
8
+ library_name: onnx
9
+ ---
10
+
11
+ # DocLayout-YOLO-DocStructBench (ONNX)
12
+
13
+ [juliozhao/DocLayout-YOLO-DocStructBench](https://huggingface.co/juliozhao/DocLayout-YOLO-DocStructBench) ์˜ ๊ฐ€์ค‘์น˜๋ฅผ **ONNX๋กœ ๋ณ€ํ™˜ํ•œ ๊ฒƒ**์ด๋‹ค.
14
+ ๊ฐ€์ค‘์น˜ ์ž์ฒด๋Š” ์›๋ณธ๊ณผ ๋™์ผํ•˜๋ฉฐ ์žฌํ•™์ŠตยทํŒŒ์ธํŠœ๋‹ํ•˜์ง€ ์•Š์•˜๋‹ค.
15
+
16
+ ## ์™œ ONNX๋กœ ๋ณ€ํ™˜ํ–ˆ๋Š”๊ฐ€
17
+
18
+ ์›๋ณธ ๊ฐ€์ค‘์น˜๋Š” Apache-2.0 ์ด์ง€๋งŒ, ์ด๋ฅผ ์‹คํ–‰ํ•˜๋Š” `doclayout_yolo` ํŒŒ์ด์ฌ ํŒจํ‚ค์ง€๋Š”
19
+ **AGPL-3.0** ์ด๋‹ค. AGPL์€ ๋„คํŠธ์›Œํฌ ์„œ๋น„์Šค๋กœ ์ œ๊ณตํ•  ๋•Œ๋„ ์†Œ์Šค ๊ณต๊ฐœ๋ฅผ ์š”๊ตฌํ•˜๋Š” ๊ฐ•ํ•œ
20
+ ์นดํ”ผ๋ ˆํ”„ํŠธ๋ผ, ์ด ํŒจํ‚ค์ง€๋ฅผ ์ œํ’ˆ์— ๋„ฃ์œผ๋ฉด ๋ผ์ด์„ ์Šค๊ฐ€ ์ „์—ผ๋œ๋‹ค.
21
+
22
+ Apache-2.0 ์ธ ๊ฐ€์ค‘์น˜๋งŒ ONNX๋กœ ๋ณ€ํ™˜ํ•ด `onnxruntime`(MIT)์œผ๋กœ ์‹คํ–‰ํ•˜๋ฉด
23
+ ๋™์ผํ•œ ์„ฑ๋Šฅ์„ ์œ ์ง€ํ•˜๋ฉด์„œ AGPL ์ฝ”๋“œ๋ฅผ ๋ฐฐํฌ๋ฌผ์—์„œ ์ œ์™ธํ•  ์ˆ˜ ์žˆ๋‹ค.
24
+
25
+ ## ์›๋ณธ ๋Œ€๋น„ ๋ณ€๊ฒฝ์‚ฌํ•ญ (Apache-2.0 ยง4 ๊ณ ์ง€)
26
+
27
+ - ์›๋ณธ: `juliozhao/DocLayout-YOLO-DocStructBench` (revision `8c3299a30b8ff29a1503c4431b035b93220f7b11`) ์˜ `doclayout_yolo_docstructbench_imgsz1024.pt`
28
+ - ๋ณ€๊ฒฝ: PyTorch `.pt` โ†’ ONNX ๊ทธ๋ž˜ํ”„๋กœ ๋‚ด๋ณด๋‚ด๊ธฐ๋งŒ ์ˆ˜ํ–‰. ๊ฐ€์ค‘์น˜ ๊ฐ’ ๋ณ€๊ฒฝ ์—†์Œ
29
+ - ๋ณ€ํ™˜ ์„ค์ •: `imgsz=1024`, `opset=17`, `simplify=False`
30
+ - ๋ณ€ํ™˜ ๋„๊ตฌ: doclayout-yolo 0.0.4, torch 2.11.0, onnx 1.22.0
31
+
32
+ ## ์ž…์ถœ๋ ฅ
33
+
34
+ | | ํ˜•ํƒœ | ์„ค๋ช… |
35
+ |---|---|---|
36
+ | ์ž…๋ ฅ `images` | `[1, 3, 1024, 1024]` float32 | letterbox(ํŒจ๋”ฉ๊ฐ’ 114) ํ›„ 0~1 ์ •๊ทœํ™”ํ•œ RGB |
37
+ | ์ถœ๋ ฅ `output0` | `[1, 300, 6]` float32 | `(x1, y1, x2, y2, confidence, class_id)` |
38
+
39
+ YOLOv10 ๊ณ„์—ด์ด๋ผ NMS๊ฐ€ ๋ชจ๋ธ ์•ˆ์— ์—†๋‹ค(NMS-free). ์ถœ๋ ฅ์ด ์ด๋ฏธ ์ตœ์ข… ๊ฒ€์ถœ ๊ฒฐ๊ณผ์ด๋ฏ€๋กœ
40
+ ํ›„์ฒ˜๋ฆฌ๋Š” ์ขŒํ‘œ ์—ญ๋ณ€ํ™˜๊ณผ ์ž„๊ณ„๊ฐ’ ํ•„ํ„ฐ๋ฟ์ด๋‹ค.
41
+
42
+ ## ํด๋ž˜์Šค (10์ข…, ์ˆœ์„œ ๊ณ ์ •)
43
+
44
+ ```
45
+ 0 title 5 table
46
+ 1 plain text 6 table_caption
47
+ 2 abandon 7 table_footnote
48
+ 3 figure 8 isolate_formula
49
+ 4 figure_caption 9 formula_caption
50
+ ```
51
+
52
+ ## ์‚ฌ์šฉ ์˜ˆ
53
+
54
+ ```python
55
+ import cv2, numpy as np, onnxruntime as ort
56
+ from PIL import Image
57
+
58
+ session = ort.InferenceSession("doclayout_yolo_docstructbench_imgsz1024.onnx",
59
+ providers=["CPUExecutionProvider"])
60
+ # letterbox ์ „์ฒ˜๋ฆฌ๋Š” ultralytics ์™€ ๋™์ผํ•ด์•ผ ์ขŒํ‘œ๊ฐ€ ๋งž๋Š”๋‹ค (cv2 INTER_LINEAR, ํŒจ๋”ฉ 114)
61
+ ```
62
+
63
+ ์ „์ฒด ๊ตฌํ˜„์€ [๊ณจ๋“ ๋งํฌ OCR ๋ชจ๋“ˆ](https://github.com/podongchip-stack/AIRookie/tree/feature/info/ocr)์˜
64
+ `src/goldenlink_ocr/layout.py` ์ฐธ๊ณ .
65
+
66
+ ## ๋ผ์ด์„ ์Šค
67
+
68
+ Apache License 2.0 โ€” ์›๋ณธ ์ €์ž‘์ž์˜ ๋ผ์ด์„ ์Šค๋ฅผ ๊ทธ๋Œ€๋กœ ๋”ฐ๋ฅธ๋‹ค.
69
+ ์›์ €์ž‘๊ถŒ์€ [juliozhao/DocLayout-YOLO-DocStructBench](https://huggingface.co/juliozhao/DocLayout-YOLO-DocStructBench) ์˜ ์ €์ž‘์ž์—๊ฒŒ ์žˆ๋‹ค.