Scene text detection and Its application on Android

This project is forked from (https://github.com/argman/EAST) Most of code is borrowed from this repository

The MobileNetV2 code is borrowed from official tensorflow mobilenet model: https://github.com/tensorflow/models/tree/master/research/slim/nets/mobilenet

Here are some changes I made:

1. Changed the architecture in model.py
2. Changed a small part of multigpu_train.py and eval.py to work with the modified architecture.
3. Add MobileNetV2 scipt and its component expanded convs to the nets folder to use MobileNetV2 and residual bottleneck layer.

For the implementation on Android:

1. Wrote a simple user interface to demo the functionality of the model.
2. Transfer the checkpoint into frozen model and configure the input and output for its implementation on Android.
3. Re-wrote some of the post-processing part including rotation, resizing and filtering from Python to Java
4. Wrote a C++/Android interface for the NMS(Non-maximam suppression)

My Android application's repository is here:

(https://github.com/ZXdatascience/TextDetectorAndroid)

My project is not suitable to display with Ipython Notebook. So there is only one Final_Report.pdf file. The report is here (https://github.com/ZXdatascience/EAST/blob/master/Final_Report.pdf)

The Android demo video is here: https://youtu.be/odfPYDGKSvM