landmarks_utils

landmarks_utils is a Python package for landmark localization model training, inference, and annotation, particularly tailored for medical imaging applications.

It builds on predict-idlab/landmarker, with a custom fork at ClemensWatzenboeck/landmarker that includes:

• A lazy DataLoader (avoids memory issues on large datasets)
• Other stability and usability fixes for large-scale inference

---

🚀 Installation (via conda)

Follow these steps to set up your environment:

# Step 1: Install CUDA (if using GPU)
# Follow official instructions from: https://developer.nvidia.com/cuda-downloads

# Step 2: Create and activate a conda environment
conda create -n landmark_env python=3.9
conda activate landmark_env

# Step 3: Clone and install the modified landmarker fork
git clone https://github.com/ClemensWatzenboeck/landmarker.git  /path/to/your/local/code/landmarker 
cd /path/to/your/local/code/landmarker
pip install -e .

# Step 4: Install this package (landmarks_utils)
cd /path/to/your/landmarks_utils
pip install -e .

How to use:

Annotation

Highlights of the provided utility functions are a useful tool to create point-annotations build with napari.

After installation and updating the example config files (see runs/config_/_annotation/config_H.yaml) it can be stated with

  landmarks_utils__annotate_landmarks  --config /path/to/runs/config/annotation/config_H.yaml

This will start the UI for the annotation.

Training landmark detection

Update the config in ./runs/config_landmarks/hands/H_train_landmarks_102p1__updated_data_handler.yaml

python ~/code/RA/landmarks_utils/landmarks_utils/training/landmarks/01_train_mlflow.py  \
   --config  `./runs/config_landmarks/hands/H_train_landmarks_102p1__updated_data_handler.yaml

Inference landmark detection

Update the config in ./runs/config_landmarks/inference/H_inference_580_cases.yaml. Insert mlflow runid of the trained model, path to data, ...

python /home/cwatzenboeck/code/RA/landmarks_utils/landmarks_utils/inference/landmarks_inference_v2.py  \
   --config  `./runs/config_landmarks/inference/H_inference_580_cases.yaml

---

📌 Annotation Examples

---

🖼️ Demo

• Annotation Demo 1: Basic point annotations
• Annotation Demo 2: Updating predictions of the model
• Annotation Demo 3: Training annotators and checking against ground truth

---

🗂️ Patch Extraction Examples

• Patch Extraction

---

📈 Metrics & Results

See notebook feet eval and notebook hands eval.

CPE Curves (F)	CPE Curves (H)

SDR (F)	SDR (H)

---