method: Naver Labs2018-06-25
Authors: Animesh Prasad, Hervé Déjean, Jean-Luc Meunier, Max Weidemann, Johannes Michael, Gundram Leifert
Description: For this task we use a pipeline approach where first the line image is preprocessed and then passed through a CNN-BLSTM architecture with CTC loss (i.e. HTR). Then in next step, we use a BLSTM over the feature layer (computed as all character n-gram for the tokens generated from best effort decoding of HTR output) trained using cross entropy loss to maximize the accuracy.
method: Joint HTR + NER no postprocessing2018-10-27
Authors: Manuel Carbonell, Mauricio Villegas, Alicia Fornés, Josep Lladós
Description: Given input lines we feed them into a CRNN model and jointly predict the transcription, named entities and person tags, combining them into an extended alphabet, predicting at each time step either a transcription symbol or the tag of the upcoming word.
Date | Method | Basic Score | Complete Score | Name | Surname | Location | Occupation | State | Input Type | |||
---|---|---|---|---|---|---|---|---|---|---|---|---|
2018-06-25 | Naver Labs | 95.46% | 95.03% | 97.01% | 92.73% | 95.03% | 96.43% | 96.41% | LINE | |||
2018-10-27 | Joint HTR + NER no postprocessing | 90.59% | 89.40% | 89.94% | 84.07% | 90.71% | 92.10% | 96.59% | LINE |