worn scroll
Would anyone be interested in contributing to an opensource dataset (of annotated license plates) to train an opensource ANPR?
The model would likely be a transformer based OCR platform trained as a MOE model to reduce inference time and reduce re-training when the dataset expands.
I cannot promise much other than an opensource repo with the dataset and if successful the model.
north tulip
Hey this seems interesting
I have been studying this repo:https://github.com/Ucas-HaoranWei/Vary.
And wanted to recreate it
worn scroll
And wanted to recreate it
I think if your trying to do that pdfs of school work and pdfs of books would be excellent as schoolwork will allow recognition of bad and good handwriting and pdfs of books will help the model generalise sentence physical structure and perfect writing.
north tulip
You can actually checkout stepfun/got-ocr-2.0 paper
Its a pretty unified e2e model that gets the job done
worn scroll
Its a pretty unified e2e model that gets the job done
This is license plate reading as an e2e model
I'm unsure on compatibility