Hi, On Monday, November 12, 2012 1:35:58 AM UTC+1, MattJ wrote:
> I'm currently training Ocropus character models, and I'm following the > example of fraktur-boxes and uw3-500. In the ocropus-align step, I've > noticed that some lines will fail the e.seg[1]==0 assertion on line 234. > Once this happens, processing stops for the remainder of the files. I've > patched this to abandon the current line and continue the for loop, but I'm > reluctant to submit a patch as I don't really follow what exactly is > happening here. The line basically says that if there is a space in the transcription, there shouldn't be a corresponding set of pixels in the segmentation. I'm not sure why this is happening, but if it happens rarely, it's probably safe to skip such lines. All you care about with alignment is to get a large amount of training data. ocropus-align implements Viterbi alignment. In the long term, we'll probably move to forward-backward training, which tends to be better behaved. OCRopus 0.7 will contain a new recognizer based on recurrent neural networks; training that is much simpler and may be a better match to your needs. Tom -- You received this message because you are subscribed to the Google Groups "ocropus" group. To post to this group, send email to [email protected]. To unsubscribe from this group, send email to [email protected]. To view this discussion on the web visit https://groups.google.com/d/msg/ocropus/-/NpDt8kFuUAoJ. For more options, visit https://groups.google.com/groups/opt_out.
