Add model, configuration files and description

#1
by mboillet - opened
Files changed (3) hide show
  1. README.md +54 -0
  2. line_hugin_munin.pth +3 -0
  3. parameters.yml +11 -0
README.md CHANGED
@@ -1,3 +1,57 @@
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
+ # Hugin-Munin line detection
18
+
19
+ The Hugin-Munin line detection model predicts text lines from Hugin-Munin document images. This model was developed during the [HUGIN-MUNIN project](https://hugin-munin-project.github.io/).
20
+
21
+ ## Model description
22
+
23
+ The model has been trained using the Doc-UFCN library on Hugin-Munin document images.
24
+ It has been trained on images with their largest dimension equal to 768 pixels, keeping the original aspect ratio.
25
+ The model predicts two classes: vertical and horizontal text lines.
26
+
27
+ ## Evaluation results
28
+
29
+ The model achieves the following results:
30
+
31
+ | set | class | IoU | F1 | AP@[.5] | AP@[.75] | AP@[.5,.95] |
32
+ | ----- | ---------- | ----- | ----- | ------- | -------- | ----------- |
33
+ | train | vertical | 88.29 | 89.67 | 71.37 | 33.26 | 36.32 |
34
+ | | horizontal | 69.81 | 81.35 | 91.73 | 36.62 | 45.67 |
35
+ | val | vertical | 73.01 | 75.13 | 46.02 | 4.99 | 15.58 |
36
+ | | horizontal | 61.65 | 75.69 | 87.98 | 11.18 | 31.55 |
37
+ | test | vertical | 78.62 | 80.03 | 59.93 | 15.90 | 24.11 |
38
+ | | horizontal | 63.59 | 76.49 | 95.93 | 24.18 | 41.45 |
39
+
40
+ ## How to use
41
+
42
+ Please refer to the Doc-UFCN library page (https://pypi.org/project/doc-ufcn/) to use this model.
43
+
44
+ # Cite us!
45
+
46
+ ```bibtex
47
+ @inproceedings{boillet2020,
48
+ author = {Boillet, Mélodie and Kermorvant, Christopher and Paquet, Thierry},
49
+ title = {{Multiple Document Datasets Pre-training Improves Text Line Detection With
50
+ Deep Neural Networks}},
51
+ booktitle = {2020 25th International Conference on Pattern Recognition (ICPR)},
52
+ year = {2021},
53
+ month = Jan,
54
+ pages = {2134-2141},
55
+ doi = {10.1109/ICPR48806.2021.9412447}
56
+ }
57
+ ```
line_hugin_munin.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8e6e5e4f272a9b43a98de036c9164fdc522d2780e2e66dd601bde930bb14a950
3
+ size 49115013
parameters.yml ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ version: 0.0.1
3
+ parameters:
4
+ mean: [209, 204, 191]
5
+ std: [51, 51, 50]
6
+ min_cc: 50
7
+ classes:
8
+ - background
9
+ - text_line_horizontal
10
+ - text_line_vertical
11
+ input_size: 768