# PBL4

**Repository Path**: feng-bu-lai/pbl4

## Basic Information

- **Project Name**: PBL4
- **Description**: No description available
- **Primary Language**: Unknown
- **License**: MulanPSL-2.0
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2026-04-14
- **Last Updated**: 2026-04-21

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# 基于轻量化特征融合和渐进式学习的工业仪表识别方法

### Industrial Meter Recognition Method Based on Lightweight Feature Fusion and Progressive Learning



## ⚙️ Requirements

Install all required dependencies into a new virtual environment via conda.

- Python == 3.9

- PyTorch ==2.5.1

- torchvision ==  0.20.1+cu121

  Install dependencies using:

  ```
  pip install -r requirements.txt
  ```

## 📊 Datasets

1. Download ADI:https://aistudio.baidu.com/datasetdetail/21521177

2. Convert the annotation format to the following format:

   ```PY
   coco_data = {
       "info": {
           "description": "OCR Crop Dataset",
           "version": "1.0"
       },
       "licenses": [],
       "categories": [
           {
               "id": 1,
               "name": "text"
           }
       ],
       "images": [],
       "annotations": [
           "id": , 
           "image_id": ,
           "category_id": ,
           "area": ,
           "bbox": ,
           "attributes": {
               "text": 
           }
       ]
   }
   ```

## 🚀 Training

1. First, generate synthetic datasets.

   ```
   python realistic_meter_synthesis.py
   ```

   

2. Configure the config.py file to specify the synthetic dataset and organize the ADI dataset into the Train\Val\Test directory structure.

3. Begin training:

   ```
   python train.py	
   ```

## 📈 Evaluation

After training, evaluate the model using:

```
python eval_sota.py --checkpoint  path/to/your/xxxxx.pth
```

Evaluation metrics include:

- Seq Acc
- Digit Seq Acc
- NED
- Char F1
- Dot F1
- Confusion Err
- Params(M)
- FPS

If you have any questions you can contact us: 3207929377@qq.com