search

Ucas-HaoranWei / GOT-OCR2.0

Official code implementation of General OCR Theory: Towards OCR-2.0 via a Unified End-to-end Model
5.67k stars 470 forks source link

Unable to extract an accuracy/confidence metric from the OCR #148

Open PantelisDeveloping opened 4 days ago

PantelisDeveloping commented 4 days ago

Dear Authors,

Thank you for this lovely project. Please i try to implement your OCR library but i am unable to implement a confidence metric of the detected OCR accuracy. If possible kindly suggest an approach or implement it. In case it has been please let me know thank you so much.

Ucas-HaoranWei commented 3 days ago

Do you mean the benchmark？ https://github.com/Ucas-HaoranWei/GOT-OCR2.0/tree/main/GOT-OCR-2.0-master/GOT/eval and you can download the benchmark in https://drive.google.com/drive/folders/1OdDtsJ8bFJYlNUzCQG4hRkUL6V-qBQaN

ayumiymk commented 3 days ago

I also have the same question: how can we obtain the probability of the OCR results?

PantelisDeveloping commented 3 days ago

Hello, thank you for your prompt response,

I don't mean the benchmarks themselves. Sometimes when I perform OCR on various documents in some cases I fail to obtain the result accurately. I would like to implement this: "For a given image we performed OCR with an accuracy of ***%". How can we measure the accuracy of the OCR ourselves for a given document image? Thank you for your understanding.

taiwanhuachenyu commented 3 days ago

希望得到类似PaddleOCR的置信度

Ucas-HaoranWei commented 3 days ago

希望得到类似PaddleOCR的置信度

那是检测的置信度吧，不是ocr（识别）的置信度吧？

Ucas-HaoranWei commented 3 days ago

Hello, thank you for your prompt response,

I don't mean the benchmarks themselves. Sometimes when I perform OCR on various documents in some cases I fail to obtain the result accurately. I would like to implement this: "For a given image we performed OCR with an accuracy of ***%". How can we measure the accuracy of the OCR ourselves for a given document image? Thank you for your understanding.

It is actually the detection accuracy？