AutoWorkHomeMatic

Inspirations

My mom is a primary school math teacher, who spent quite a lot of time marking worksheets from about 80 students. It would make her and all other teachers like her have an easier life if computers can do some work for them. There are commercial softwares that tell whether elementary arithmetic questions are answered correctly, but that is all they can do. Also they depend on cell phones, which are not productive. Therefore, I came up with the idea that machine learning can be integrated with the efficiency of printers and scanners, in order to judge and mark answers on worksheets quickly.

What I learnt

1. Python image process libraries: opencv pillow
2. Python virtual environment
3. operating GNU/Linux
4. Git undo a push to Github

How is it built

Please go to the github readme page for detailed description.

Challenges

GNU/Linux

This is my first time developing a project with GNU/Linux as the operating system. I was using Windows for all of my previous projects and I started switching to Ubuntu 20.04 recently.

Tensorflow with CUDA

I have a gaming PC with GTX 1070. At the beinging, I intends to run tensorflow with it, since traning a network with GPU can be much faster than CPU, however, I ran into driver, dependency and compatibiliy issues. I decided that I shall not waste all my time messing up with the environment. Eventually it turns out to be a wise decision, since a pre-trained model is found and there is not time left for collecting training set and perform training.

Built With

• keras
• opencv
• pillow
• pytesseract
• python
• tensorflow
• yolo