Code formatted by autopep8.
Requires Python version 3.12.6 for compatibility with Tensorflow.
Dependencies can be installed by executing this command from the src folder of the project:
pip install -r requirements.txt
The model can be run by executing this command from the src folder of the project:
python3 main.py
Alternatively, the epoch info can be viewed directly on GitHub without running locally in the Actions tab under the run_model.yml workflow.
Predicted mask visually matches reasonably well with true mask.
Tracked metrics from 20 epochs are:
- Accuracy 88.67%
- Loss 27.82
- Mean IoU 0.42
Overall accuracy and especially mean IoU would likely improve with additional images and iterations, as the current sample of 20 is relatively small.
This project uses the Surgical Scene Segmentation in Robotic Gastrectomy Kaggle dataset.
The project proposal, presentation, and report can be found in the docs folder.