[ECCV24] Bones Can't Be Triangles: Accurate and Efficient Vertebrae Keypoint Estimation through Collaborative Error Revision

[Project Page]

Introduction

This is the official PyTorch implementation of "Bones Can't Be Triangles: Accurate and Efficient Vertebrae Keypoint Estimation through Collaborative Error Revision (ECCV 2024)."

Environment Setup

This code was developed using Python 3.8 on an Ubuntu 18.04 system.

Quick start

Installation

1. Install Required Packages: Use pip to install the necessary Python packages from the requirements.txt file:

   pip install -r requirements.txt

2. Data Preparation:

   • Obtain the dataset: The AASCE dataset can be requested from this link.

   • Organize the dataset: Move the downloaded dataset to the following directory structure:

     codes/preprocess_data/AASCE_rawdata/boostnet_labeldata
     

   • Run preprocessing: Navigate to the preprocessing code directory and execute the preprocessing script:

     cd codes/preprocess_data/
     python preprocess_data.py
     cd ..

How to use

1. Training Your Own Model:

   To train your model, execute the following command:

   bash train_interactive_keypoint_model.sh
   python train_AASCE.py
   

2. Inference:

   Once the data is prepared, run the following command to perform inference with the pre-trained model:

   python evaluate_AASCE.py
   

Citation

If you find this work or code is helpful in your research, please cite:

@inproceedings{kim2024Bones,
  title={Bones Can't Be Triangles: Accurate and Efficient Vertebrae Keypoint Estimation through Collaborative Error Revision},
  author={Kim, Jinhee and Kim, Taesung and Choo, Jaegul},
  booktitle={European Conference on Computer Vision},
  year={2024},
}