Skip to content

Swap out EasyOCR model for a better trained one #8

@ckcollab

Description

@ckcollab

Trained all night on my GPU and this is the best I got with my old dataset with lots of error prone hand-done data + newly generated one...

[300000/300000] Train loss: 0.00381, Valid loss: 0.00855, Elapsed_time: 50642.84251
Current_accuracy : 98.454, Current_norm_ED  : 0.9977
Best_accuracy    : 98.551, Best_norm_ED     : 0.9982

I think that's maybe 1% better than the old model, will add a couple tests where this model is proven to outperform the last..

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions