Building a desktop activity classifier for Mac by training a Convolutional Neural Network to distinguish between desktop screen activities using transfer learning. The final model achieves an accuracy of 97.24% on the test set.
The project report and model results can be found at https://github.com/m15e/capstone-computer-vision-time-tracker/blob/master/capstone_project.pdf
Please note that the first version of this application unfortunately only runs on Mac
- Python 3.5
- Pytorch
- fastai
- pyautogui
- pync
All files are contained in the timeNet.
- The project report can be found under timeNet/mlnd_capstone.pdf
- The test data label mapping can be found in timeNet/test.csv
- The files for training the benchmark and final models can be found under timeNet/benchmark_model.ipynb and timeNet/classifier_training.ipynb
- Open the terminal
- Navigate to the timenet folder
- Run
python timeNet.py