search

qurator-spk / eynollah

Document Layout Analysis
Apache License 2.0
332 stars 27 forks source link

Multi class Instance Segmentation #41

Closed ss756 closed 1 year ago

ss756 commented 3 years ago

Hi, I would love to test out your pre-trained models on some documents. Could you point me to the right model to perform pixel wise region (text, image, separator) classification. Would be of gr8 help. Keep up the amazing work!!

vahidrezanezhad commented 3 years ago

Dear Suyash,

Eynollah uses a mixture of models based on the input and options wished by the user. Eynollah by default can detect main texts, separators, images, background, and marginals. If you are interested in headings and drop capitals you need to apply -fl option. Based on your source test dataset (if they are very dark or bright ) you can turn -ib (--input_binary) option. As you see based on the dataset and user-case Eynollah uses different models. You can share some of your documents and I may help you how you can use Eynollah to get optimum results :)

cneud commented 3 years ago

See also https://github.com/qurator-spk/eynollah#models.

cneud commented 3 years ago

We should also come up with better names and introduce versioning for the individual models @vahidrezanezhad - cf. https://github.com/qurator-spk/eynollah/issues/5.

cneud commented 3 years ago

The models and their names should then be shortly described in the models section of the README.