blahblahhhj / onr

Tensorflow CNN on MNIST dataset with B/S interface

ONR

Tensorflow CNN on MNIST dataset with B/S interface for handwrite number recognition

Requirements

• MNIST dataset

• numpy==1.13.3

• tensorflow==1.2.1

• (If you have tensorflow-gpu, then cuDNN and CUDA are needed. Be careful of version conflict:))

• os

• time

• json

• http

• struct

  Use the requirements.txt by typing this at your terminal:

  pip install -r requirements.txt
  

How to use?

1. Download MNIST dataset (4 files)
2. Unzip all 4 files and put them into MNIST_Data folder (parallel to this README) in the project.
3. Run server.py.

• (If it is the first time for running, training the model will take less than ten minutes.)

4. After Server Ready is printed, open the browser and go to localhost:8000.
5. Draw you digit (from 0 to 9) in the canvas, and press recognize.
6. Wait for the result to show.

• Might not give you the right answer, especially when you draw number 6:)
• Have fun!

Model detail

Tensorboard for training

Tensorboard for predicting