Digit recognition in Python using TensorFlow.
This tool tries to analyse image files containing a digit, and prints out what digit it is.
It's programmed in Python, and uses TensorFlow.
The image resolotuion for the custom testing data is 28x28 pixels.
The installation commands bellow, assume that you're using Bash. If you're using PowerShell/Command Prompt (kekw), replace all instances of python3 with py in the commands.
- Install Python (3.8 used in development)
- Install TensorFlow (
python3 -m pip install tensorflow) - Install OpenCV (
python3 -m pip install opencv-python) - Install NumPy (
python3 -m pip install numpy)
You have to train the model before using it, run python3 train.py
Then, to see how the model perceieves the testing data in the testing/ directory, run python3 main.py
You should see results with the filename, and the answers coming on screen. Do note the model isn't perfect, but has a really good accuracy. When I tested it, I once managed to get an 90% success rate when I used the model against the testing set that I had made.
If you want to use your own images, draw numbers on a 28x28 image, and place them in the testing/ directory, and name the files using numbers, in ascending order,
starting from 1. So your file names should be 1.png, 2.png, 3.png, and so on.
On running the train.py file, we see the progress in training the model
A folder, with the model name is then created, where the model is stored.
On running the main.py file, we see the model predicting the digits present in the testing/ directory
