---
language:
  - ru
license: mit
tags:
  - ocr
  - easyocr
  - text-recognition
  - classification
  - lightgbm
  - russian
  - apex-legends
  - custom-trained
datasets:
  - custom
library_name: easyocr
model-index:
  - name: Cyrillic OCR Fine-Tuned (EasyOCR)
    results: []
  - name: Player Tier Classifier (LightGBM)
    results: []
---

# 🛠 Models for Apex-Stat-Detector-Discord-Bot

This repository contains **two custom-trained models** developed specifically for the [AI-Apex-Stat-Detector-Discord-Bot](https://github.com/PSImera/AI-Apex-Stat-Detector-Discord-Bot.git):

1. 📄 **OCR Model (EasyOCR)** — for extracting Cyrillic stats text from Apex Legends profile screenshots  
2. 🧠 **Tier Classifier (LightGBM)** — for evaluating player skill level based on recognized stats

---

## 🔤 OCR Model – Cyrillic Text Recognition (EasyOCR)

This model is based on `cyrillic_g2.pth`, fine-tuned to recognize the font used in the Apex Legends in-game statistics page, which contains both **Cyrillic** and **Latin** characters.

It is designed to extract information such as **damage**, **K/D**, **win rate**, and other numeric indicators from game profile screenshots.

### 🖼 Sample Inputs

![damage](images/damage-annot.png)
![kd](images/kd-annot.png)

---

### 🔧 OCR Model Info

- **Framework**: EasyOCR  
- **Base**: `cyrillic_g2.pth`  
- **Character Set**: Cyrillic + Latin
- **Format**: `.pth`  
- **Trained on**: Custom annotated dataset 
- **Purpose**: Enhanced recognition of Cyrillic text in low-resolution images with stylized fonts.

---

### 🚀 OCR Usage Example

```python
import easyocr

reader = easyocr.Reader(
    ["en", "ru"],
    model_storage_directory="models/EasyOCR_model",
    user_network_directory="models/EasyOCR_user_network",
    recog_network="apex_stats_detector",
)

results = reader.readtext("image.jpg")
print(results)
```