Robust ReadingCompetition

method: Super Fast Scene Text Detector_v22018-06-01

Authors: HIT Graduation Project

Description: A CNN based model. It is fast but with little compensation on accuracy.
12.5 FPS on 720 * 1280 (720P) images and 33 FPS on 512 * 512 images, which outperforms most of current state-of-the-art methods.
F-measure 84.4 is got from the model Pretrained on COCOtext 2014 and finetuned on ICDAR2015 training data (without ensemble and multi-scale testing).
With only ICDAR 2015 training data, this model's F-measure is 82.5.
We believe our model can boost current FOTS model a lot by replacing its detection module.