Results

The 2013 edition of the competition is now on at http://dag.cvc.uab.es/icdar2013competition. Please consider submitting your results to the new edition of the competition.

Note: Please note that despite its name, ground truth is never perfect. Our ground truth (for both training and test) contains 522 images and more than 4500 word bounding boxes. With all the feedback we received after the ICDAR 2011 competition we have corrected/added about 20 word bounding boxes (a ~0.5% of the total). It is imperative to always take this ground truthing error margin into account when doing comparisons based on the results. This is especially true when examining results at the level of individual images, which are only provided for qualitative checks, and are not meant to be indicative of the overall performance. All results given below are calculated with the latest version of the ground truth (v2.0) which was produced after the ICDAR competition, taking into account the feedback we have received. This is not planned to change again, so consider it a stable version. If you are interested in seeing results based on the v1.0 of the ground truth (used for the ICDAR 2011 competition) please see this page.

You can see below the latest results, updated in real time when new submissions are made to the system. Methods published in ICDAR 2011 in grey, plus other public methods in white and your methods in yellow.

Ranking for Task 1 - Text Localization

ParticipantMethodRecallPrecisionHmean
XuwangYinTexStar84.21 %93.49 %88.61 %details..
Yang HuanTCC_textDetector88.64 %81.46 %84.90 %details..
Marios Anthimopoulos 10.1007/s10044-011-0237-781.88 %87.35 %84.53 %details..
yanghaojinVOCR80.66 %83.37 %82.00 %details..
kaolin fireBlindsight201272.93 %89.58 %80.40 %details..
Chao ZengCTD74.88 %85.35 %79.78 %details..
AlvaroAlvaro Gonzalez70.08 %89.23 %78.51 %details..
Yunxue ShaoTDM_IACAS69.70 %85.83 %76.93 %details..
Chen YangTH-TextLoc73.06 %80.39 %76.55 %details..
samsTextorter69.08 %85.54 %76.43 %details..
BaselineBaseline69.94 %83.92 %76.30 %details..
Deepak KumarOTCYMIST75.65 %63.85 %69.25 %details..
cyiSASA64.91 %67.38 %66.12 %details..
Muhammad Shehzad HanifText Hunter58.43 %75.52 %65.88 %details..

Ranking for Task 2 - Text Segmentation

ParticipantMethodWell s.MergedLostRecallPrecisionHmean
XuwangYinTexStar75.96 %9.64 %14.41 %87.22 %79.18 %83.01 %details..
yanghaojinVOCR_adaptive_skeleton68.73 %15.70 %15.56 %84.14 %77.54 %80.71 %details..
yanghaojinVOCR_global_skeleton68.81 %15.68 %15.50 %84.04 %77.57 %80.68 %details..
MariosAnthimopoulos72.98 %12.18 %14.83 %86.15 %71.19 %77.96 %details..
Deepak KumarOTCYMIST64.14 %15.69 %20.15 %80.62 %72.06 %76.10 %details..
xinhao liuAdaptiveEdgeDetection66.55 %9.24 %24.20 %78.24 %70.97 %74.43 %details..
samsTextorter58.12 %9.50 %32.37 %65.23 %63.63 %64.42 %details..
cyiSASA41.58 %10.97 %47.43 %71.68 %55.44 %62.52 %details..

Ranking for Task 3 - Word Recognition

ParticipantMethodTotal Edit distanceCorrectly Recognised Words
yanghaojinVOCR106.6 82.03 %details..
Deepak KumarPower-law Transformation for Enhanced Recognition108.7 82.9 %details..
yanghaojinVOCR_2145.4 74.95 %details..
MinhHieu280513186.6 72.88 %details..
Bolan Suproposed_abbyy188.2 72.77 %details..
Chen YangTH-OCR189.1 61.98 %details..
AlvaroAlvaro Gonzalez226.8 66.88 %details..
BaselineBaseline231.2 63.94 %details..
Important Dates
  • 25 February: Webpage is now online! Registrations page is open.
  • 20 March: Training dataset will be available
  • 1 June: Test dataset available
  • 1-3 June: Test period
  • 3 June: Submission of results