Skip to main content

Submission

The real world is full of surprises, and you're bound to encounter situations where things don't quite fit.

Our model is no different; it may not handle every situation perfectly.

If you find instances where our model doesn't perform as expected, we encourage you to provide us with datasets. We'll use the datasets you provide to adjust and optimize the model accordingly.

We sincerely appreciate your willingness to contribute datasets, and we'll test and integrate them as soon as possible.

Format Guidelines

For this task, you need to provide text templates without any margins or backgrounds, similar to the datasets we have open-sourced.

You don't need to provide any labels; we'll automatically label your dataset during integration.

We recommend uploading the data to your Google Drive and providing us with the link via email. Upon receiving your data, we'll test it promptly. If your provided data doesn't meet our requirements, we'll inform you promptly.

  • Reasons for data rejection may include:
    • Insufficient dataset accuracy: For example, your dataset contains images with background noise rather than full-sized, calibrated images.
    • Low dataset resolution: Although we trained the model using 128 x 128 images, we expect the original image quality not to be too poor— at least recognizable to the human eye. We believe this requirement is already quite low, and we apologize if it causes you any inconvenience.

Frequently Asked Questions

  1. Will submitting my data guarantee better performance?

    • Not necessarily. While exposing the model to your provided data is beneficial, it doesn't guarantee significant improvements. It's better to have seen the data than not, but it doesn't always lead to substantial enhancements.
  2. How important are file names?

    • File names are not a primary concern; they only need to correctly link to the corresponding images.
  3. Any recommendations for image formats?

    • We recommend using the JPG format to save space.

Contact Us

For further assistance, please contact us via email: docsaidlab@gmail.com