Update README.md
Browse files
README.md
CHANGED
|
@@ -11,15 +11,21 @@ language:
|
|
| 11 |
- sv
|
| 12 |
pipeline_tag: object-detection
|
| 13 |
---
|
|
|
|
| 14 |
# Historical Document Layout Detection Model
|
|
|
|
| 15 |
A fine-tuned Mask R-CNN model (via LayoutParser/Detectron2) for detecting layout
|
| 16 |
elements in historical Swedish medical journal pages.
|
|
|
|
| 17 |
## Model Details
|
|
|
|
| 18 |
- **Model type:** Mask R-CNN (ResNet backbone)
|
| 19 |
- **Framework:** Detectron2 / LayoutParser
|
| 20 |
- **Fine-tuned for:** Historical document layout analysis
|
| 21 |
- **Language of source documents:** Swedish
|
|
|
|
| 22 |
## Label Map
|
|
|
|
| 23 |
| ID | Label |
|
| 24 |
|----|------------------|
|
| 25 |
| 0 | Advertisement |
|
|
@@ -31,21 +37,30 @@ elements in historical Swedish medical journal pages.
|
|
| 31 |
| 6 | Table |
|
| 32 |
| 7 | Text |
|
| 33 |
| 8 | Title |
|
|
|
|
| 34 |
## Usage
|
|
|
|
| 35 |
### Installation
|
|
|
|
| 36 |
Follow instructions at:
|
| 37 |
https://detectron2.readthedocs.io/en/latest/tutorials/install.html
|
|
|
|
| 38 |
### Finetuning
|
| 39 |
-
|
|
|
|
| 40 |
https://detectron2.readthedocs.io/en/latest/tutorials/training.html
|
|
|
|
| 41 |
### Inference
|
|
|
|
| 42 |
```python
|
| 43 |
import cv2
|
| 44 |
import layoutparser as lp
|
| 45 |
import matplotlib.pyplot as plt
|
|
|
|
| 46 |
# Configuration
|
| 47 |
model_config_path = "config_mask_rcnn_resized.yaml"
|
| 48 |
model_path = "model_final_LP.pth"
|
|
|
|
| 49 |
label_map = {
|
| 50 |
0: "advertisement",
|
| 51 |
1: "author",
|
|
@@ -57,6 +72,7 @@ label_map = {
|
|
| 57 |
7: "text",
|
| 58 |
8: "title",
|
| 59 |
}
|
|
|
|
| 60 |
# Load model
|
| 61 |
model = lp.models.Detectron2LayoutModel(
|
| 62 |
config_path=model_config_path,
|
|
@@ -64,18 +80,31 @@ model = lp.models.Detectron2LayoutModel(
|
|
| 64 |
extra_config=["MODEL.ROI_HEADS.SCORE_THRESH_TEST", 0.8],
|
| 65 |
label_map=label_map,
|
| 66 |
)
|
|
|
|
| 67 |
# Load and process image
|
| 68 |
image = cv2.imread("<path_to_image>")
|
| 69 |
image = image[..., ::-1] # BGR to RGB
|
|
|
|
| 70 |
# Detect layout
|
| 71 |
layout = model.detect(image)
|
|
|
|
| 72 |
# Print detected elements
|
| 73 |
for block in layout:
|
| 74 |
print(f"Type: {block.type}, Score: {block.score:.3f}, Box: {block.coordinates}")
|
|
|
|
| 75 |
# Visualize results
|
| 76 |
viz = lp.draw_box(image, layout, box_width=3, show_element_type=True)
|
| 77 |
plt.figure(figsize=(12, 16))
|
| 78 |
plt.imshow(viz)
|
| 79 |
plt.axis("off")
|
| 80 |
plt.show()
|
| 81 |
-
```
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 11 |
- sv
|
| 12 |
pipeline_tag: object-detection
|
| 13 |
---
|
| 14 |
+
|
| 15 |
# Historical Document Layout Detection Model
|
| 16 |
+
|
| 17 |
A fine-tuned Mask R-CNN model (via LayoutParser/Detectron2) for detecting layout
|
| 18 |
elements in historical Swedish medical journal pages.
|
| 19 |
+
|
| 20 |
## Model Details
|
| 21 |
+
|
| 22 |
- **Model type:** Mask R-CNN (ResNet backbone)
|
| 23 |
- **Framework:** Detectron2 / LayoutParser
|
| 24 |
- **Fine-tuned for:** Historical document layout analysis
|
| 25 |
- **Language of source documents:** Swedish
|
| 26 |
+
|
| 27 |
## Label Map
|
| 28 |
+
|
| 29 |
| ID | Label |
|
| 30 |
|----|------------------|
|
| 31 |
| 0 | Advertisement |
|
|
|
|
| 37 |
| 6 | Table |
|
| 38 |
| 7 | Text |
|
| 39 |
| 8 | Title |
|
| 40 |
+
|
| 41 |
## Usage
|
| 42 |
+
|
| 43 |
### Installation
|
| 44 |
+
|
| 45 |
Follow instructions at:
|
| 46 |
https://detectron2.readthedocs.io/en/latest/tutorials/install.html
|
| 47 |
+
|
| 48 |
### Finetuning
|
| 49 |
+
|
| 50 |
+
Follow instructions at:
|
| 51 |
https://detectron2.readthedocs.io/en/latest/tutorials/training.html
|
| 52 |
+
|
| 53 |
### Inference
|
| 54 |
+
|
| 55 |
```python
|
| 56 |
import cv2
|
| 57 |
import layoutparser as lp
|
| 58 |
import matplotlib.pyplot as plt
|
| 59 |
+
|
| 60 |
# Configuration
|
| 61 |
model_config_path = "config_mask_rcnn_resized.yaml"
|
| 62 |
model_path = "model_final_LP.pth"
|
| 63 |
+
|
| 64 |
label_map = {
|
| 65 |
0: "advertisement",
|
| 66 |
1: "author",
|
|
|
|
| 72 |
7: "text",
|
| 73 |
8: "title",
|
| 74 |
}
|
| 75 |
+
|
| 76 |
# Load model
|
| 77 |
model = lp.models.Detectron2LayoutModel(
|
| 78 |
config_path=model_config_path,
|
|
|
|
| 80 |
extra_config=["MODEL.ROI_HEADS.SCORE_THRESH_TEST", 0.8],
|
| 81 |
label_map=label_map,
|
| 82 |
)
|
| 83 |
+
|
| 84 |
# Load and process image
|
| 85 |
image = cv2.imread("<path_to_image>")
|
| 86 |
image = image[..., ::-1] # BGR to RGB
|
| 87 |
+
|
| 88 |
# Detect layout
|
| 89 |
layout = model.detect(image)
|
| 90 |
+
|
| 91 |
# Print detected elements
|
| 92 |
for block in layout:
|
| 93 |
print(f"Type: {block.type}, Score: {block.score:.3f}, Box: {block.coordinates}")
|
| 94 |
+
|
| 95 |
# Visualize results
|
| 96 |
viz = lp.draw_box(image, layout, box_width=3, show_element_type=True)
|
| 97 |
plt.figure(figsize=(12, 16))
|
| 98 |
plt.imshow(viz)
|
| 99 |
plt.axis("off")
|
| 100 |
plt.show()
|
| 101 |
+
```
|
| 102 |
+
|
| 103 |
+
## Acknowledgements
|
| 104 |
+
|
| 105 |
+
This model builds upon the excellent work of:
|
| 106 |
+
|
| 107 |
+
- [Detectron2](https://github.com/facebookresearch/detectron2/tree/main)
|
| 108 |
+
- [LayoutParser](https://github.com/Layout-Parser/layout-parser?tab=readme-ov-file)
|
| 109 |
+
|
| 110 |
+
We thank the contributors and maintainers of these projects for making their tools publicly available and supporting research.
|