File size: 1,032 Bytes
aa3ec1e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
---
license: mit
base_model:
- Ultralytics/YOLOv8
pipeline_tag: object-detection
---
# Overview
This repository hosts a YOLOv8l model trained on the ArxivFormula (https://github.com/microsoft/ArxivFormula) dataset, which focuses on the detection of mathematical expressions in scientific papers.

# Training Data:
- Source: ArxivFormula (https://github.com/microsoft/ArxivFormula)
- Classes: 6 classes (InlineFormula, DisplayedFormulaLine, FormulaNumber, DisplayedFormulaBlock, Table, Figure)
  Pages: ~600,000 images of document pages

# Model:
- YOLOv8l (https://github.com/ultralytics/ultralytics)
- epochs = 100
- imgsz = 640
- optimizer = 'AdamW'
- lr0 = 0.0001
- augment = True

# Usage
## Example Code

```
from ultralytics import YOLO
import pathlib

# Sample images
img_list = ['sample1.png', 'sample2.png', 'sample3.png']

# Load the document segmentation model
model = YOLO('arxivFormula_YOLOv8l.pt')

# Process the images
results = model(source=img_list, save=True, show_labels=True, show_conf=True, show_boxes=True)
```