README.md
PrePareded
- anaconda env
- requirements.txt
- clova AI → deep text recognition → trained weights (ex, .pth)
- wpod-net weights (ex, .h5 , .json)
- https://github.com/KavenLee/wpod_ocr/releases/tag/wpod
- Make Dir ‘weights’ and put the downloaded file in the Dir
OCR Recognition Training
- train data, validation data (ex, image)
- train & val labelling data (ex, .txt)
- train & val data split using ocr_label.py & train_test_split.py
- Label (ex, {imagepath} \t {label} \n ) → a.jpg(\t)apple(\n)
- cd d:\Library\deep-text-recognition-benchmark
- using train.py, create-lmdb.py
Detect Number Plate
- using wpod-net
- get high accuracy if only a car existed in image file.
- using openCV image
- using PIL Image on OCR
Excution
- using a image
- success
- Failed
- using video
Performance
- No GPU, Only CPU (OS : Windows)
- Detect Number Plate time → Iamge : 100 ~ 180ms ,Video : 200ms
- OCR Recognition time → Image : 50ms , Video : 80 ~ 100 ms
- Total time → 200~300ms
- Using GPU on Jetson Tx2 (OS : Linux)
- Detect Number Plate time → Image : 3 ~ 400ms
- OCR Recognition time → Image : 100~ 200ms
- Total time → 400~600 ms
Reference
https://github.com/clovaai/deep-text-recognition-benchmark
네이버 deep-text-recognition 모델을 custom data로 학습 & 아키텍쳐 분석
