|
|
--- |
|
|
license: apache-2.0 |
|
|
language: |
|
|
- en |
|
|
metrics: |
|
|
- accuracy |
|
|
base_model: |
|
|
- Inventic-AI/Account_Number_Extracter |
|
|
pipeline_tag: image-to-text |
|
|
tags: |
|
|
- finance |
|
|
--- |
|
|
|
|
|
## Description |
|
|
|
|
|
This repository provides three YOLO-based models intended to be used sequentially to extract and recognize digits from document images. |
|
|
|
|
|
The pipeline works in three stages: |
|
|
|
|
|
1. **Region Segmentation** |
|
|
|
|
|
* `segmenter.pt` (YOLOv11n) |
|
|
* Finds the account number region in a document image |
|
|
* Output is cropped and passed to the next stage |
|
|
|
|
|
2. **Digit Detection** |
|
|
|
|
|
* `BBox.pt` (YOLOv11n) |
|
|
* Detects bounding boxes for each digit within the cropped region |
|
|
* Bounding boxes should be sorted left-to-right |
|
|
|
|
|
3. **Digit Classification** |
|
|
|
|
|
* `Classify.pt` (YOLOv11s-cls) |
|
|
* Classifies each cropped digit image into labels `0–9` |
|
|
* Predictions are concatenated to form the final sequence |
|
|
|
|
|
### Sample Output: |
|
|
Stage 1: |
|
|
|
|
|
 |
|
|
|
|
|
Stage 2: |
|
|
|
|
|
|
|
|
 |
|
|
|
|
|
--- |
|
|
|
