Accuracy with Hindi and English text s not up to mark
Get following output
G-CHARGE
India's Fastest Growing Brand
5 Kg नींबू
10 Kg आटा
250g चीरा
1 Expert Set
250g घासपत्ती
250g सींग
500g भारतीय दाल
500g देवेशी दाल
500g चीज
500g चीज पत्ती
1 dettol on Set
5 kg Vita Ghee
5 Kg चावड
500 मुंडी मसरी बीन दाल
APPROVED when i upload follwing image
Hi @ravi19 , thanks for reporting this issue and sharing the example.
We reproduced a similar setup during internal testing and found that when the same text content is re-rendered as clean printed text, the model can recognize the Hindi–English mixed content accurately. This suggests that the current errors are mainly related to handwriting-style appearance and visual variability, rather than a limitation in multilingual (Hindi + English) recognition itself.
At the moment, recognition performance on handwriting and handwriting-like text is still weaker than on printed text, and improving handwriting robustness is one of our current optimization priorities. We’re actively working on expanding training coverage for such scenarios.
Thanks again for your feedback!