Add model, configuration files and description

#1
by mboillet - opened
Files changed (3) hide show
  1. README.md +75 -0
  2. model.pth +3 -0
  3. parameters.yml +11 -0
README.md CHANGED
@@ -1,3 +1,78 @@
1
  ---
 
2
  license: mit
 
 
 
 
 
 
 
 
 
 
3
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
+ library_name: Doc-UFCN
3
  license: mit
4
+ tags:
5
+ - Doc-UFCN
6
+ - PyTorch
7
+ - Object detection
8
+ metrics:
9
+ - IoU
10
+ - F1
11
+ - AP@.5
12
+ - AP@.75
13
+ - AP@[.5,.95]
14
  ---
15
+
16
+
17
+ # Generic historical line detection
18
+
19
+ The generic historical line detection model predicts text lines from document images.
20
+
21
+ ## Model description
22
+
23
+ The model has been trained using the Doc-UFCN library on 10 historical document datasets including these public datasets:
24
+ * [Bozen](https://zenodo.org/record/218236)
25
+ * [cBAD2017 (READ)](https://zenodo.org/record/1491441)
26
+ * [cBAD2019](https://zenodo.org/record/2567398)
27
+ * [DIVA-HisDB](https://diuf.unifr.ch/main/hisdoc/diva-hisdb.html)
28
+ * [Horae](https://github.com/oriflamms/HORAE/)
29
+ * [ScribbleLens](https://www.openslr.org/84/)
30
+
31
+ It has been trained on images with their largest dimension equal to 768 pixels, keeping the original aspect ratio.
32
+
33
+ ## Evaluation results
34
+
35
+ The model achieves the following results on the test sets:
36
+
37
+ | | IoU | F1 | AP@[.5] | AP@[.75] | AP@[.5,.95] |
38
+ | ----------------------- | ----- | ----- | ------- | -------- | ----------- |
39
+ | Bozen | 60.15 | 75.10 | 97.14 | 3.79 | 27.50 |
40
+ | cBAD2017 (READ) Complex | 46.79 | 60.35 | 56.01 | 3.40 | 16.26 |
41
+ | cBAD2017 (READ) Simple | 53.97 | 68.43 | 57.26 | 8.45 | 19.39 |
42
+ | cBAD2019 | 50.77 | 64.52 | 35.46 | 2.88 | 11.51 |
43
+ | DIVA-HisDB | 41.54 | 57.88 | 63.15 | 0.00 | 11.69 |
44
+ | Horae | 48.93 | 63.95 | 57.45 | 5.20 | 15.55 |
45
+ | ScribbleLens | 76.61 | 86.72 | 98.02 | 71.87 | 58.32 |
46
+
47
+ The model has been trained to reduce mergers in predictions (see the [paper](https://link.springer.com/article/10.1007/s10032-022-00395-7) for more details on training). Therefore, despite slightly low evaluation values, the model correctly detects lines on a wide variety of historical and modern manuscript documents.
48
+
49
+ ## How to use
50
+
51
+ Please refer to the Doc-UFCN library page (https://pypi.org/project/doc-ufcn/) to use this model.
52
+
53
+ # Cite us!
54
+
55
+ ```bibtex
56
+ @inproceedings{boillet2022,
57
+ author = {Boillet, Mélodie and Kermorvant, Christopher and Paquet, Thierry},
58
+ title = {{Robust Text Line Detection in Historical Documents: Learning and Evaluation Methods}},
59
+ booktitle = {{International Journal on Document Analysis and Recognition (IJDAR)}},
60
+ year = {2022},
61
+ month = Mar,
62
+ pages = {1433-2825},
63
+ doi = {10.1007/s10032-022-00395-7}
64
+ }
65
+ ```
66
+
67
+ ```bibtex
68
+ @inproceedings{boillet2020,
69
+ author = {Boillet, Mélodie and Kermorvant, Christopher and Paquet, Thierry},
70
+ title = {{Multiple Document Datasets Pre-training Improves Text Line Detection With
71
+ Deep Neural Networks}},
72
+ booktitle = {2020 25th International Conference on Pattern Recognition (ICPR)},
73
+ year = {2021},
74
+ month = Jan,
75
+ pages = {2134-2141},
76
+ doi = {10.1109/ICPR48806.2021.9412447}
77
+ }
78
+ ```
model.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b5d0659a29638446f573bec8928f3380f8643f4bee5c9ea7b8e64ae813dcc6f5
3
+ size 49198561
parameters.yml ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ version: 0.0.1
3
+ parameters:
4
+ mean: [194, 185, 160]
5
+ std: [49, 49, 47]
6
+ min_cc: 50
7
+ classes:
8
+ - background
9
+ - text_line
10
+ input_size: 768
11
+