FAST_TextProposal

• Combination of Text Proposals algorithm with Fully Convolutional Networks to efficiently reduce the number of proposals while maintaining the same recall level.
  

Installation

• This code requires:

  1. OpenCV 3.1 (tested with 02edfc8)
  2. Caffe (tested with d21772c)
  3. tinyXML

• Installation:

  • cd EarlyPruning
  • cmake .
  • make

Run

• Steps of generating proposals:

  1. Heatmaps: in order to generate the heatmaps, you can train your model and save your hatmaps according to this link.

  2. Early pruning: run the shell command for generating the proposals:
     for i in {1..500}; do sh -c "echo 'Processing $i' && ./img2hierarchy /path/to/input/img_${i}.jpg /path/to/trained_boost_groups.xml /path/to/heatmap/img_${i}.png 0.14 > /path/to/proposals/img_$i.csv 2>/dev/null"; done

Evaluation

• Computing the confidences
  confIoU.py prop2conf ./proposals/*.csv -threads=10
• Computing the IoU
  confIoU.py conf2IoU ./conf_proposals/*.csv -threads=10
• Plot the detection rate
  confIoU.py '-extraPlotDirs={".":"proposals"}' getCumRecall ./conf_proposals/img_* '-IoUThresholds=[0.5]' -maxProposalsIoU=100000 -care=1

Citation

Please cite this work in your publications if it helps your research:
@article{Bazazian17,
author = {Bazazian, Dena and Gomez, Raul and Nicolaou, Anguelos and Gomez, Lluis and Karatzas, Dimosthenis and Bagdanov, Andrew D.},
title = {FAST: Facilitated and Accurate Scene Text Proposals through FCN Guided Pruning},
journal = {Pattern Recognition Letters(2017)},
year = {2017},
ee = {doi: 10.1016/j.patrec.2017.08.030 }
}