unilm icon indicating copy to clipboard operation
unilm copied to clipboard

Published 20 hours ago • verified publisher icon microsoft

Layoutlmv3 for RE

Open SuXuping opened this issue 2 years ago • 3 comments

Describe when i use layoutlmv3 to do RE task on XFUND_zh dataset, the result is 'eval_precision': 0.5283, 'eval_recall': 0.4392. i do not konw the reason of the bad result. maybe there is something wrong with my RE task code? maybe i need more data for training? is there some suggestions for me to improve the result?
Dose anyone meet the same problem?

SuXuping avatar Oct 25 '22 10:10 SuXuping

hi @SuXuping , I think it is a good idea to reproduce the score of layoutlmv3-Chinese on the SER task first to make sure you use the layoutlmv3-Chinese rightly. If there is any further questions, feel free to ask me : )

Dod-o avatar Nov 17 '22 04:11 Dod-o

Hi, Can you please share the code?

ShingHuuu avatar Nov 30 '22 06:11 ShingHuuu

hi @SuXuping , I think it is a good idea to reproduce the score of layoutlmv3-Chinese on the SER task first to make sure you use the layoutlmv3-Chinese rightly. If there is any further questions, feel free to ask me : )

Hi，i have done SER task with LayoutlmV3 and the result is pretty good. The result is: f1score:0.9096 but i still cannot improve RE task. Have you trained the RE task? what about your result?

SuXuping avatar Dec 05 '22 03:12 SuXuping