analysis.jl
- Notebook that loads all of the various .csv files from the
/data/*folder and prepares figures and tables for analysis and publication
benchmarks.ipynb
- Google colab notebook that benchmarks our GPU accelerated python-wrapper distance transform against various Python-based distance transforms
benchmarks.jl
- Notebook that benchnmarks all of Julia distance transforms against ours (2D and 3D)
- Maurer (ImageMorphology.jl)
- Felzenswalb Single-Threaded CPU (DistanceTransforms.jl)
- Felzenswalb Multi-Threaded CPU (DistanceTransforms.jl)
- Proposed CUDA (DistanceTransforms.jl)
- Proposed AMDGPU (DistanceTransforms.jl)
- Proposed Metal (DistanceTransforms.jl)
hausdorff_loss_timings.ipynb [In Progress]
- Google colab notebook that benchmarks our DT in the hausdorff loss vs. other Python-based DTs vs. pure dice loss and benchmarks full training loops of the various hausdorff loss functions
hausdorff_loss_accuracy.ipynb [In Progress]
- Google colab notebook that provides full training loop of pure dice vs hausdorff + dice loss accuracy comparisons
- Outputs quantitative date (dice, hausdorff, iou, etc.) and qualitative images showing the different models
skeletonization.jl [TODO]
- Notebook that implements a GPU based skeletonization algorithm (using our GPU distance transform) and compares against a pure CPU skeletonization algorithm (using a CPU distance transform)
/data
- Contains the various timings and accuracy data as .csv files from the above notebooks
/plots
- Contains .png files created from the
analysis.jlnotebook for publications