TH Model Update Schedule and Case Sensitivity Issue in OCR

[iey704]

I'm wondering when the TH model will be updated.

Also, when I perform OCR using the combination of the en_PP-OCRv3_det_infer detection model and the latin_PP-OCRv3_rec_infer recognition model, it doesn't distinguish between uppercase and lowercase letters. Could you provide an explanation for this?

[GreatV]

Regarding your questions:

1. TH Model Update Schedule: There is no mention of a schedule for updating the TH model in the provided discussions. Typically, model updates depend on contributions from the community and the development priorities of the PaddleOCR team. You may check the official PaddleOCR repository for announcements or updates regarding new model releases.

2. Case Sensitivity Issue in OCR:

   • Your issue with case sensitivity when using the en_PP-OCRv3_det_infer detection model and the latin_PP-OCRv3_rec_infer recognition model is likely due to the character dictionary used in the recognition model. Some recognition models in PaddleOCR only support lowercase by default.
   • The character dictionary (character_dict_path) used by the recognition model determines whether it can distinguish between uppercase and lowercase letters. If case sensitivity is critical for your use case, you may need to modify the dictionary that the model uses and retrain or fine-tune it accordingly.
   • A similar issue is discussed in this PaddleOCR Discussion: Fine-tuning recognition model en_PP-OCRv4_rec, where modifying the character dictionary is suggested as an option during fine-tuning of the recognition model.

Recommendations:

• Modify Character Dictionary: Check if the latin_PP-OCRv3_rec_infer model supports case-sensitive recognition by examining the character dictionary it uses (typically found in ppocr/utils/latin_dict.txt or related files). If uppercase letters are missing, you can modify the dictionary to include both uppercase and lowercase characters.
• Fine-Tune the Recognition Model: If modifying the dictionary alone does not work, fine-tuning the model with a dataset that includes case-sensitive character labels could improve recognition performance.
• Check Model Configuration: Ensure that your model configuration does not normalize text to lowercase during pre-processing or post-processing.

Would you like more detailed instructions on modifying the character dictionary or fine-tuning the model?

---

Response generated by 🤖 feifei-bot | chatgpt-4o-latest

⚠️ This is an AI-generated response. Please verify any critical information.

[iey704]

Please tell me about detailed instructions on modifying the character dictionary or fine-tuning the model.
Thank you.