We will compare the performance of OCR and DTRB and see how accurate the results are in correctly identifying Iranian license plates
pip install -r requirements.txt
Give the language you are considering to easyOCR and see the results with the Matplotlib library. You can remove the percentage of detection probabilities with a simple Python condition and see those that have a higher percentage of probability.
More complete description and use of EasyOCR
More complete description and use of deep-text-recognition-benchmark
Download via IR-LPR repository
clone the DTRB repository, We run the demo.py file that was previously taught for a few pictures to make sure of the correctness of our work so far, and we download the file with the pth extension, which is already designed and trained with the architecture, and we use the transfer training method. We use it to teach Iranian license plates that have not been seen on the network.
--train_data dataset/train --valid_data dataset/validation \
--select_data / --batch_ratio 1 --num_iter 10000 --batch_max_length 8 --valInterval 50 \
--Transformation TPS --FeatureExtraction ResNet --SequenceModeling BiLSTM --Prediction Attn
Image name | 00005.jpg | 00008.jpg | 01290.jpg | 01666.jpg | 01309.jpg | 01310.jpg |
---|---|---|---|---|---|---|
Image | ||||||
Recognized text from image | 95t38633 | 44h67355 | 42t7711 | 73i34174 | 27t54314 | 37t58922 |
Letters | ت | ه | ت | ی | ت | ت |