ECS171-FinalProject - Team 14

Topic: Letter recognition
PRoject Report: report

Demo Instruction

Frontend Webpage

Located under Demo folder

• DemoLocal.html should be ready to open with any browser but need to run the backend server to be able to utilize our model and recived prediction.
• DemoInternet.html is under Demo/templates, is same website but hosted by server
  • (still only work for local unless you set up the router fowrding, behaves the same)

Backend Server

Located under "Demo" folder

• backend.py file hosts the server, install all needed library below and use command "python backend.py" in Demo folder to run
• routes_predicts.py file contains the prediction routes/functions for the server to handel the request of using one version of our model.
• Libary needed
  • flask & flask-cors for hosting server using python, intall with "pip intsall flask" & "pip intsall flask-cors"
  • PIL for image processing and save local copy of collected image, intall with "pip intsall pillow"
  • numpy and tensorflow for data processing and model calling

Also, change line 257 of DemoLocal.html const IP_address = "http://127.0.0.1:5000" to the correct address that server is running at. Example terminal output when run server: =====================

PS ........ECS171-FinalProject\demo> python .\backend.py

* Serving Flask app 'backend'<\br>

WARNING: This is a development server. Do not use it in a production deployment. Use a production WSGI server instead.

* Running on all addresses (0.0.0.0)

* Running on http://127.0.0.1:5000

=============================================================

Model Instruction

All model training code locates under ModelTraining folder

• Final Data Preprocessing.ipynb collects all the preprocessing functions we created.
• Final Model.ipynb contains the training process and the performance analysis graphs.
• CNN_ver8_grid_search.ipynb contains the grid search histroy for the last hyper parameter tunning step, and it generated the structure of our final model.
• The rest notbook files which named with version number are the cleaned-up version of recorded attemps on data-processing / hyper parameter tuning with some analysis.