Skip to content

Aurora11111/text-detection-train-yolov3

 
 

Darknet Logo

Darknet

Darknet is an open source neural network framework written in C and CUDA. It is fast, easy to install, and supports CPU and GPU computation.

For more information see the Darknet project website.

For questions or issues please use the Google Group.

some using detaile can be found in below link:

https://github.com/AlexeyAB/darknet

text-detection-yolov3

this project is used to detect text mainly reference:

https://github.com/Aurora11111/keras_ocr

https://github.com/tianzhi0549/CTPN

https://github.com/pjreddie/darknet.git

datasets prepare

first, you can down load icdar2017:链接:http://rrc.cvc.uab.es/?ch=7&com=downloads

then :python voc_annotation.py

if your dataset format is json:

python dict_datasets.py

attentions 1:the accordibnate sequences must be:(xmin, xmax,ymin, ymax)

generate anchors

python kmeans.py

attentions 1:the accordibnate sequences must be:(xmin, ymin, xmax, ymax)

attentions 2:file structural must be similarity to voc

training

./darknet detector train cfg/voc.data cfg/text.cfg darknet53.conv.74

Releases

No releases published

Packages

No packages published

Languages

  • C 88.8%
  • Cuda 7.7%
  • Python 2.3%
  • Other 1.2%