table_structure_recognition icon indicating copy to clipboard operation
table_structure_recognition copied to clipboard

Published 20 hours ago • verified publisher icon whn09

Questions about training environment

Open nhathoangx007 opened this issue 2 years ago • 3 comments

Thanks for your repo. I have some questions about your training environment, in order to reproduce the result.

  1. How much RAM needed for training yolov5 with this large dataset?
  2. Which GPU(s) did you use?
  3. How long did the training last?

I am currently working on the field of table recognition and your repo is very interesting. Thank you very much.

Edit: Mentioned @whn09 for reference.

nhathoangx007 avatar Feb 08 '23 20:02 nhathoangx007

Hi,

I trained the models for about 10 hours/model on AWS ml.p3.2xlarge (1 NVIDIA V100 GPU (16GB GPU Memory), 8 CPU, 64GB RAM).

whn09 avatar Feb 17 '23 02:02 whn09

@whn09 Thanks, may I ask one more question? What is your recommendation about OCR for postprocessing?

nhathoangx007 avatar Feb 24 '23 04:02 nhathoangx007

I think PaddleOCR is suitable for English and Chinese, and Amazon Textract or other SaaS services may be better.

whn09 avatar Mar 05 '23 09:03 whn09