Bonnet: Ultra-Fast Whole-Body Bone Segmentation from CT Scans

Bonnet is an ultra-fast whole-body bone segmentation pipeline for CT scans. It runs in seconds per scan on a single commodity GPU while maintaining reliable segmentation quality across different datasets.

---

Contents

• Links: Processed training data & model weights
• Training
• Inference with sample data

---

Links: Processed training data & model weights

• Processed training data:
  Processed Data

• Model weights:
  Weights

---

Training

1. Open the configs:

   • Bonnet/conf/config_eva.yaml
   • Bonnet/conf/data/totalseg_hu200_3000.yaml

2. Set paths and options:

   • In Bonnet/conf/config_eva.yaml, configure your output/log paths and other training options. Make sure the dataset selection points to the correct data config you are using.
   • In Bonnet/conf/data/totalseg_hu200_3000.yaml, set the local paths for dataset_path and cache_path to match your machine.

3. Run training:

python main.py

Inference with sample data

This repo includes sample data for inference. You only need to:

1. point the data config to the correct local path.
2. download weights and point the main config to the checkpoint.

Step 1: Point the data config to the sample data

1. Open:
   • Bonnet/conf/data/totalseg_hu200_3000.yaml
2. Set dataset_path and cache_path to the correct local path of the sample data in this GitHub repo.

Step 2: Download weights and set checkpoint path

1. Download the model checkpoint from:
   • Weights
2. Put the downloaded checkpoint under the Bonnet/ directory.
3. Open:
   • Bonnet/conf/config_eva.yaml
4. Update the checkpoint field in this config (e.g., checkpoint_path / root_path / checkpoints_dir, depending on your config) to the correct local path of the downloaded weight file.

Step 3: Enable eval-only mode and run

1. Open:
   • Bonnet/conf/eval/eval_on_test.yaml
2. Set:

eval_only: True

3. Run:

python main.py