SenNetworks

Submission for Kaggle competition Hacking the Human Vasculature in 3D

Required packages

• Numpy
• cv2
• matplotlib
• torch
• torchvision
• pandas
• sklearn
• tqdm

Running UNET

The sample data conting 5 training images and 5 test images is already present. Only need to run the unet.ipynb notebook.

Running Segformer + Segment Anything (Assuming just testing)

1. Ensure that you have the data folder with 6 images and 6 labels at the inner-most level
2. Enable your conda environment or python virtual env, open transformers.ipynb

Segformer:

1. Run the packages cell
2. Run the dataset cell
3. Run the 'ingest data' cell, shape should be (6,) for both
4. Run the split cell
5. Run the dataset / dataloader cell
6. Optional: Run sanity check cell
7. Run custom loss cell
8. Run Segformer custom cell
9. Scroll down to model load cell (after save cell), run it
10. Run decode cell
11. Run test cell, should observe results below

SAM:

1. Run all SAM cells, observe output at bottom