Skip to content

Question about the Training Procedure for the SVTRv2 Chinese Recognition Model #209

Description

@yuyuyu864

Hello, I would like to ask about the complete training procedure for the SVTRv2 Chinese text recognition model.
In the readme, the English model is trained in two stages: first, svtrv2_rctc.yml is used to train the first-stage model and obtain ./output/rec/u14m_filter/svtrv2_rctc/best.pth; then, svtrv2_smtr_gtc_rctc.yml loads this checkpoint for the second-stage training.Does the Chinese model also require two-stage training? If so, which configuration file should be used for the first stage? Should I first train a Chinese RCTC/CTC model on the BCTR training set, or can I directly load the English checkpoint ./output/rec/u14m_filter/svtrv2_rctc/best.pth?
Thank you very much!

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