https://github.com/manish7294/regional-language-detector
Detecting the location and native language of a place from an image
https://github.com/manish7294/regional-language-detector
crnn deep-learning image-processing image-recognition language-classification neural-networks object-detection opencv tensorflow
Last synced: 3 months ago
JSON representation
Detecting the location and native language of a place from an image
- Host: GitHub
- URL: https://github.com/manish7294/regional-language-detector
- Owner: manish7294
- License: mit
- Created: 2017-10-28T03:44:53.000Z (almost 8 years ago)
- Default Branch: master
- Last Pushed: 2017-10-30T03:08:50.000Z (almost 8 years ago)
- Last Synced: 2025-07-19T17:08:15.397Z (3 months ago)
- Topics: crnn, deep-learning, image-processing, image-recognition, language-classification, neural-networks, object-detection, opencv, tensorflow
- Language: Python
- Homepage:
- Size: 26.9 MB
- Stars: 0
- Watchers: 1
- Forks: 2
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Regional-Language-Detector
A regional language and location detecting application using a tri-combination of Google's TensorFlow Object Detection API, Convolutional Recurrent Neural Network (CRNN) and a 2-layered Neural Network for language classification.
The Application uses three modules in sequential manner:
1. Google's TensorFlow Object Detection API for Text localisation
2. CRNN for text Recognition
3. 2-layered Neural Network for language classification
*Location is determined by predicted text using Geotext*
### Starting Application
- *Edit the Image path in main.py*
```sh
#Image-Location-Detector/
python3 main.py
```
### Requirements
- [Python 3.6]
- [TensorFlow 1.2]
- [PyTorch]
- [OpenCv] - *Either build from source or*
```sh
sudo apt-get install python-opencv
```
* [Geotext]
```sh
pip3 install geotext
```
### Downloads
wget https://www.dropbox.com/s/l0vo83hmvv2aipn/crnn.pth
### Notes
- Currently `ssd_mobilenet_v1_coco_11_06_2017` model is being used - *boundry box prediction for text detection will be poor*
- Try to build your own custom dataset using `labelImg`
### Todos
- Train custom dataset for Text detection
- [Notes on generating custom dataset](http://androidkt.com/train-object-detection/)
### Refrences
- https://github.com/AKSHAYUBHAT/DeepVideoAnalytics/tree/master/notebooks/OCR
- https://github.com/datitran/object_detector_app
