Udacity MLND Capstone Project

---

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

Dependencies

---

Please note that the first version of this application unfortunately only runs on Mac

• Python 3.5
• Pytorch
• fastai
• pyautogui
• pync

Files

---

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

Usage

---

1. Open the terminal
2. Navigate to the timenet folder
3. Run python timeNet.py