cnn_lstm_ctc_lpr

KO | EN

gitlite — search

#javascript #python #hacktoberfest #react #ai #typescript #llm #go #golang #android #machine-learning #rust #deep-learning #linux

cnn_lstm_ctc_lpr

★ 8 Open GitHub ↗

cnn + lstm + ctc for unconstrained lpr

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

cnn + lstm + ctc for unconstrained lpr
=========
[![All Contributors](https://img.shields.io/badge/all_contributors-1-blue.svg?style=flat-square)](https://raw.githubusercontent.com/alvenchen/cnn_lstm_ctc_lpr/master/)

- This repo contains code and demo for chinese licence plate recognition(lpr) using cnn and lstm.

- This repo is heavily inspired by [HyperLPR](https://github.com/zeusees/HyperLPR), increased accuracy in wild environment.

- Please star if it's useful for you.

Feature
---------
- end to end, no need to split word

- in the wild, accuracy is around 96.6%

TODO
---------
- multi-line recognition

- more trainning data, especially for oblique data

- licence plate detection accuracy using yolo3/SSD with more data.

Data
---------
I did't found much useful data for both detection and recognition, except [功能车牌图像库数据](http://www.openits.cn/openData/index.jhtml)

We will thankful if you share some infomation.

Train
---------
```bash
python main.py --mode train --train_dir 'your_trainning_data_dir' --val_dir 'validation_dir'
```

Test
---------
```bash
python main.py --mode infer --infer_dir 'your_test_data_dir' --restore 
```

API
---------
```bash
python api.py 
```

API Test
---------
```bash
curl -F "image=@/Users/alvenchen/Downloads/lp1.jpg"  http://127.0.0.1:8088/PlateRec
```

Result
---------
![image](https://raw.githubusercontent.com/alvenchen/cnn_lstm_ctc_lpr/master/images/result.png)

![image](https://raw.githubusercontent.com/alvenchen/cnn_lstm_ctc_lpr/master/images/dark.jpg)

![image](https://raw.githubusercontent.com/alvenchen/cnn_lstm_ctc_lpr/master/images/皖A92141.jpg)

cnn_lstm_ctc_lpr

Explore Similar Repositories

Was this content helpful?

Recent Feedback

Download README