search

PaddlePaddle / PaddleOCR

Awesome multilingual OCR toolkits based on PaddlePaddle (practical ultra lightweight OCR system, support 80+ languages recognition, provide data annotation and synthesis tools, support training and deployment among server, mobile, embedded and IoT devices)
https://paddlepaddle.github.io/PaddleOCR/
Apache License 2.0
44.6k stars 7.85k forks source link

PicoDet training on Custom Layout Dataset #8697

Closed bely66 closed 1 year ago

bely66 commented 1 year ago

Hi Everyone, I'm training the layout model using a custom dataset for Layout with only Tables and Text. I downloaded the CDLA model as a pretrained model to fine-tune on my data. The dataset has 1200 document in a non-latin and non-chinese language. The model didn't cross 20% accuracy till now. What might be the issue? Do I need to train for longer number of epcochs I set it to 100 initially? Do I need to add more data? What can be done to help the model generalize?

an1018 commented 1 year ago

You can try use the pretrained model of PubLayNet