|
--- |
|
license: mit |
|
metrics: |
|
- mean_iou |
|
datasets: |
|
- Riksarkivet/placeholder_region_segmentation |
|
tags: |
|
- mmdet |
|
- htrflow_core |
|
- instance segmentation |
|
library_name: htrflow |
|
inference: false |
|
pipeline_tag: image-segmentation |
|
--- |
|
|
|
## Model Description |
|
|
|
**RTMDet** is both an instance segmentation and object detection model from [OpenMMLab](https://mmyolo.readthedocs.io/en/latest/recommended_topics/algorithm_descriptions/rtmdet_description.html) and was trained using [MMDetection](https://mmdetection.readthedocs.io/en/latest/). This RTMDet model is fine-tuned to segment text regions within the documents, which enables a pre-localization text-line regions, which is a crucial step for current text-recognition models work at the text-line level. |
|
|
|
## Usage |
|
|
|
```python |
|
#WIP |
|
``` |
|
|
|
## Evaluation |
|
|
|
(WIP) |
|
|
|
## Training Data |
|
|
|
(WIP) |
|
|
|
## References |
|
|
|
If you would like to learn more about the Swedish National Archives HTR pipeline or access the training data, please refer to the following resources: |
|
|
|
- [The AI-lab at the Swedish National Archives](https://github.com/Swedish-National-Archives-AI-lab) |
|
- [MMDetection](https://github.com/open-mmlab/mmdetection) |
|
- [RTMDET Paper](https://paperswithcode.com/paper/rtmdet-an-empirical-study-of-designing-real) |
|
|
