ICFHR2018 Competition on Automated Text Recognition on a READ Dataset


All submissions and results for the ICFHR2018 Competition on Automated Text Recognition on a READ Dataset

Name Method Info Submitter Affiliation Result is public Track Subtrack CER
US-UC3M_DA - The CNN-BLSTM-CTC architecture used in https://arxiv.org/pdf/1804.01527.pdf at ICFHR 2018 Conference, trained without any LM. Data Augmentation applied. José Carlos Aradillas Universidad de Sevilla and Universidad Carlos III de Madrid Track Subtrack 1 0.17391851458292554
US-UC3M - The CNN-BLSTM-CTC architecture used in https://arxiv.org/pdf/1804.01527.pdf at ICFHR 2018 Conference, trained without any LM. José Carlos Aradillas Universidad de Sevilla and Universidad Carlos III de Madrid Track Subtrack 1 0.18931237940773468
finalResultsLITIS OCR conv blstm ctc + data augmentation (inclination, scaling) with LM based on multigrams with interpolation scheme. Data augmentation in test + ROVER algorithm LITIS Laboratoire d’Informatique, du Traitement de l’Information et des Systèmes, France, Yann Soullard Laboratoire d’Informatique, du Traitement de l’Information et des Systèmes, France Track Subtrack 1 0.14458138754509103
RESWLM Results with generic LM and then adapted. Joan Andreu Sanchez Universitat Politècnica de València Track Subtrack 1 0.21540924470792203
UOB-PTECH-BASELINE00 UOB and ParisTech baseline system ParisTech Telecom ParisTech, France, and University of Bala- mand, Lebanon, Chafic Mokbel, Edgard Chammas Telecom ParisTech, France, and University of Bala- mand, Lebanon, University of Balamand Track Subtrack 1 0.20937613601409358
11LMI conv 11 LMI raw LITIS Laboratoire d’Informatique, du Traitement de l’Information et des Systèmes, France, Yann Soullard Laboratoire d’Informatique, du Traitement de l’Information et des Systèmes, France Track Subtrack 1 0.21508416990576326
MDLSTM MDLSTM fed with grayscale images and a simple language model. Research Group in Pattern Recognition and Digital Image Processing (RPPDI), Byron L. D. Bezerra University of Pernambuco, Research Group in Pattern Recognition and Digital Image Processing (RPPDI), University of Pernambuco, Research Group in Pattern Recognition and Digital Image Processing (RPPDI) Track Subtrack 1 0.273251586924303
OSU_Submission_8 A CNN-RNN model was trained with CTC. Data were augmented by both grid distortion (Wigington et al 2017) and rescaling, rotation and shearing. A model was trained on the general data before fine-tuning the whole model on the respective specific datasets with test-side augmentation. Oliver Nina, Russell Ault OSU Track Subtrack 1 0.17856044294063364
test5 test Mohamed Yousef Bassyouni Faculty of Computers and Information, Assiut University, Egypt Track testtrack
AU_gctc_1 - A CTC-trained network with custom architecture. - Staged elastic and perspective distortions were used to augment images. - Images were used as-is without the provided mask. - CTC-greedy decoding is used directly without LM or any other tricks A paper will present all the details of the method Mohamed Yousef Bassyouni Faculty of Computers and Information, Assiut University, Egypt Track Subtrack 1 0.134210452727832
UOB-PTECH-BASELINE02 UOB and ParisTech baseline system - No LM ParisTech Telecom ParisTech, France, and University of Bala- mand, Lebanon, Chafic Mokbel, Edgard Chammas Telecom ParisTech, France, and University of Bala- mand, Lebanon, University of Balamand Track Subtrack 1 0.19242331031011436
AU_gctc_2 A deeper version of the network used in AU_gctc_1 Mohamed Yousef Bassyouni Faculty of Computers and Information, Assiut University, Egypt Track Subtrack 1 0.13019770140656023
FinalOCRLM OCR convolutional + BLSTM, fine-tuned with data augmentation, stride adaptation, and using cross validation. Language Model interpolating between a writer LM and a language-based LM. Perplexity measure for language detection and for selecting the interpolation parameter. ROVER on test. LITIS Laboratoire d’Informatique, du Traitement de l’Information et des Systèmes, France, Yann Soullard Laboratoire d’Informatique, du Traitement de l’Information et des Systèmes, France Track Subtrack 1 0.188054025334862
  • 13 items

News

May 16, 2018:
The competition remains open beyond the ICFHR deadline. Feel free to submit new methods.

April 15, 2018:
test data available

January 22, 2018:
competition is open and training data available

Important Dates

January 22, 2018:
competition opens

January 22, 2018:
training data available

April 15, 2018:
test data available

May 1, 2018:
deadline for submitting results on the test data

May 16, 2018:
provide a brief system description

August 5-8, 2018:
Results announced at ICFHR 2018





Organizers







Tobias Strauß

[University of Rostock, CITlab – Computational Intelligence Technology Lab] 

Gundram Leifert

[Computational Intelligence Technology Laboratory / CITlab, Rostock, Germany] 

READ Partner

Tobias Hodel

[Staatsarchiv des Kantons Zürich] 

Researcher in Digital Humanities and Digital History.