Text Detection in natural images using a Fully Convolutional Network developed during my Stage experience for my Computer Science B.Sc thesis.
Spaghetti code warning. I do not expect anyone to try and run this so:
- If you're looking for a maintainable and easily deployable FCN implementation, please refer to TensorFCN.
- If you really want to go through all of this, please refer to Chapter 4 of my thesis and/or take a look at the code of all the dataset readers and the final stages for bounding boxes extraction for each ICDAR competition submission.
For results in the various competitions check out: