Baseline + polygon detection of handwriting

[stevethearkiv]

Hello. I have ran P2Pala succesfully against typewritten and print with the default model, but I am not getting very good results when running against handwriting. Is it possible to train a model to work against handwriting, and if so, what kind of ground truth, and how much is required?

[ghost]

P2PaLa is created for handwriting text in mind, but it can somewhat generalize to printed text.

• post sample of your detection, training log, number of images used in training, sample of images trained vs trying to detect, the training config, are you talking about baselines or polygons, etc..
  Without such info, how can we diagnose your issue
  @stevethearkiv

[stevethearkiv]

Maybe I need to explain better. I run the default ALAR model against handwriting where the handwriting deviates a bit from a straight line. The rectangle will cut the top and bottom off the handwriting, and where the handwriting starts to dip, it will partially disappear out of the rectangle. So my question is. Is it possible to train a model (using the P2PaLa set of tools), to create polygons around irregular, non fixed-height handwriting. I have many images, and many sets of coordinates which define where the handwriting is; coords which follow the shape of the letters. Is this all I need, and how could I train such a model, if possible?

[ghost]

here you go dear
https://github.com/PRHLT/pageDistanceBasedContourGenerator
https://github.com/PRHLT/pageLineExtractor

[lquirosd]

Hi,
@stevethearkiv , P2PaLA did not support complex text-line extraction. It is designed to baseline detection in mind, while text-line extraction is leaved to another tools like the ones pointed out in the previous comment.