Awesome

D2GPLand [MICCAI'24][Oral]

Official Implementation of MICCAI-2024 Oral paper "Depth-Driven Geometric Prompt Learning for Laparoscopic Liver Landmark Detection".

Jialun Pei, Ruize Cui, Yaoqian Li, Weixin Si, Jing Qin, and Pheng-Ann Heng

👀 [Paper]; [Official Version]

Contact: wx.si@siat.ac.cn, peijialun@gmail.com

🔧 Environment preparation

The code is tested on python 3.9.19, pytorch 2.0.1, and CUDA 11.7, change the versions below to your desired ones.

1. Clone repository:

git clone https://github.com/PJLallen/D2GPLand.git

cd D2GPLand

2. Set up anaconda environment:

# Create D2GPLand anaconda environment from YAML.file
conda env create -f D2GPLand.yaml
# Activate environment
conda activate D2GPLand

📈 Dataset preparation

💥 Download proposed L3D dataset

• L3D dataset: Google Drive

Register datasets

Change the path of the datasets as:

DATASET_ROOT = 'D2GPLand/L3D/'
TRAIN_PATH = os.path.join(DATASET_ROOT, 'Train/')
TEST_PATH = os.path.join(DATASET_ROOT, 'Test/')
VAL_PATH = os.path.join(DATASET_ROOT, 'Val/')

🚀 Pre-trained weights

D2GPLand with SAM-b and ResNet-34: Google Drive

⚙️ Usage

Train

python train.py --data_path {PATH_TO_DATASET} \
--batch_size 4 --lr 1e-4 --decay_lr 1e-6 --epoch 60

Please replace {PATH_TO_DATASET} to your own dataset dir

Eval

python test.py --model_path {PATH_TO_THE_MODEL_WEIGHTS} \
  --prototype_path {PATH_TO_THE_PROTOTYPE_WEIGHTS} \
  --data_path {PATH_TO_DATASET}

• {PATH_TO_THE_MODEL_WEIGHTS}: please put the pre-trained model weights here
• {PATH_TO_THE_PROTOTYPE_WEIGHTS}: please put the pre-trained prototype weights here
• {PATH_TO_DATASET}: please put the dataset dir here

Acknowledgement

This work is based on:

• AdelaiDepth
• Segment Anything Model

Thanks them for their great work!

📚 Citation

If this helps you, please cite this work:

@inproceedings{pei2024land,
  title={Depth-Driven Geometric Prompt Learning for Laparoscopic Liver Landmark Detection},
  author={Pei, Jialun and Cui, Ruize and Li, Yaoqian and Si, Weixin and Qin, Jing and Heng, Pheng-Ann},
  booktitle={MICCAI},
  year={2024},
  organization={Springer}
}