Training Custom Plates Dataset

[soheer]

Hello,

We have a dataset consists of street images containing shop plates with different fonts, sizes, language and styles, and we want to use EasyOCR for plates detection. for that, we want to enhance the result, so we want to train our custom dataset.

1- And as you mentioned in "How to train your custom model" section, the training method depends on "TextRecognitionDataGenerator", but this method doesn't suits our needs. Is there a way for training other than text recognition (based on our images), and if so, what are the steps (tool for labeling images,.....), and how we can transfer it into model?

2- Can we use two models (default model and our custom model) in EasyOCR detection at the same time?

Thanks in advance.

[masoudMZB]

1- You can edit "TextRecognitionDataGenerator" code that generates images like your test data distribution. this is super easy and the code is really clean. If you have your own dataset use label studio as a tool to label your recognition and detection dataset

2- Why do you need this? you can create two Easyocr objects in your code and load 2 models, I didn't get your use case.