ketanmore's picture
Update README.md
21f72c4 verified
|
raw
history blame
1.81 kB
---
license: mit
---
![alt text](https://encrypted-tbn0.gstatic.com/images?q=tbn:ANd9GcSksxbjFqxppkfVHAN30x6JjNc_3JGeGILZPA&s "Title")
# Fine Tuning Script For Layout Model Of Surya OCR.
This repository contains [layout-fine-tune.ipynb](https://huggingface.co/ketanmore/surya-layout-fine-tune-script/blob/main/layout-fine-tune.ipynb) file, Please use this file to fine tune [Surya Layout Model](https://huggingface.co/vikp/surya_layout2). This model uses modified architecture of Segformer.
## Setup Instructions
### Clone the Surya OCR GitHub Repository
```bash
git clone https://github.com/vikp/surya.git
cd surya
```
### Switch to v0.4.14
```bash
git checkout f7c6c04
```
### Install Dependencies
You can install the required dependencies using the following command:
```bash
pip install -r requirements.txt
```
# Image Pre-processing
For image pre-processing we can directly import a function and image processor from [surya ocr github repository](https://github.com/VikParuchuri/surya/tree/v0.4.14).
```python
from surya.input.processing import prepare_image_detection
```
```python
from surya.model.detection.segformer import load_processor
```
```python
from PIL import Image
image = Image.open("path/to/image")
images = [prepare_image_detection(img=image, processor=load_processor())]
```
```python
import torch
images = torch.stack(images, dim=0).to(model.dtype).to(model.device)
```
# Loading Model
```python
from surya.model.detection.segformer import load_model
```
```python
model = load_model("vikp/surya_layout2")
```
```python
output = model(pixel_values=images)
```
### Note : Loss function
[Surya-layout-Model](https://huggingface.co/vikp/surya_layout2) does not have pre-defined loss function, We have to define it according to our dataset and the Requirements.