H3DE-Net - GitLite

KO | EN

gitlite — search

#redis #cpp #containers #tailwindcss #ruby #git #api #postgresql #ai-agent #ai

H3DE-Net

(★ 15)

No description available.

파일 탐색기

ZIP 다운로드

- - profiles_settings.xml
- .gitignore
- code.iml
- misc.xml
- modules.xml
- workspace.xml
- - __init__.cpython-310.pyc
  - __init__.cpython-36.pyc
  - __init__.cpython-37.pyc
  - __init__.cpython-39.pyc
  - dataloader.cpython-310.pyc
  - dataloader.cpython-36.pyc
  - dataloader.cpython-37.pyc
  - dataloader.cpython-39.pyc
  - transforms.cpython-310.pyc
  - transforms.cpython-37.pyc
- __init__.py
- dataloader.py
- transforms.py
- - __init__.cpython-310.pyc
  - __init__.cpython-37.pyc
  - Biformer_Unet.cpython-310.pyc
  - BiFormer_Unet1.cpython-310.pyc
  - Biformer_Unet3.cpython-310.pyc
  - losses.cpython-310.pyc
  - losses.cpython-37.pyc
  - PBiFormer_Unet1.cpython-310.pyc
  - VNet.cpython-310.pyc
  - VNet.cpython-37.pyc
- - _common.py
  - bra_legacy.py
- __init__.py
- BiFormer_Unet.py
- losses.py
- PBiFormer_Unet.py
main_baseline.py
main_yolo_saveFile.py
main_yolol.py
README.md
utils.py

# CDN으로 사용하기

jsDelivr

jsDelivr는 공개 GitHub 리포지토리를 별도 설정 없이 CDN으로 즉시 서빙합니다. 버전과 파일을 고르면 웹페이지에 바로 붙일 수 있는 링크와 예시 코드가 만들어집니다.

버전

파일

링크

예시

명령어 용어집

이 문서에서 사용된 명령어를 모아봤습니다. 낯선 명령어가 있다면 펼쳐서 확인해보세요.

🔍

`python`

설명 보기 ▼

Python 언어 인터프리터.
더 많은 정보: <https://docs.python.org/using/cmdline.html>.

python

REPL(대화형 셸) 시작:

python {{path/to/file.py}}

특정 Python 파일 실행:

python -i {{path/to/file.py}}

특정 Python 파일 실행 후 REPL 시작:

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

## 1. Introduction
### 1.1 What for?

The project of **3D BiFormer-Net** aims to accurately localize anatomical landmarks in 3D medical images, which is a critical task for applications such as surgical navigation, disease diagnosis, and treatment planning.

### 1.2 HighLights
* **3D BiFormer-Net** is designed to meet the demands of clinical environments, providing robust and accurate landmark detection even in challenging scenarios such as low-resolution images, occlusions, or missing landmarks.
* We propose a hybrid framework that combines the strengths of **CNNs** and **Transformers**. CNNs are used for efficient local feature extraction, while the Transformer module, equipped with a **bi-level routing attention mechanism**, enhances global context modeling. This combination addresses the limitations of standalone architectures.
* Extensive experiments on the **[Head Landmark_24nc]** dataset demonstrate that **3D BiFormer-Net** significantly outperforms existing baselines across multiple metrics, particularly in scenarios involving complex anatomical structures and missing landmarks.

## 2. Preparation
### 2.1 Requirements
- python >=3.7
- pytorch >=1.10.0
- Cuda 10 or higher
- numpy
- pandas
- scipy
- nrrd
- time

### 2.2 Data Preparation
<!The dataset will be available soon!>
The dataset is available at https://drive.google.com/file/d/1NGsBbqXZLDlkiSJtDQdyMlXzgnkFoVON/view?usp=sharing>
* Data division
```
    - mmld_dataset/train     # 458 samples for training
    - mmld_dataset/val       # 100 samples for validation
    - mmld_dataset/test      # 100 samples for testing
```
* Data format
```
    - *_volume.nrrd     # 3D volumes
    - *_label.npy       # landmarks
    - *_spacing.npy     # CT spacings, used for calculating MRE
```

## 3. Train and Test
### 3.1 Training baseline heatmap regression model

```
python main_baseline.py --model_name BiFormer_Unet   # network training for baseline heatmap regression model using backbone BiFormer_Unet
```

### 3.2 Training with PBiFormer_Unet

```
python main_yolol.py --model_name PBiFormer_Unet              # network training using backbone PBiFormer_Unet
```

### 3.3 Fine-tuning in a pretrained checkpoint

```
python main_baseline.py --resume ../SavePath/baseline/model.ckpt

python main_yolol.py --resume ../SavePath/yolol/model.ckpt
```

### 3.4 Metric counting
```
python main_baseline.py --test_flag 0 --resume ../SavePath/baseline/model.ckpt  # calculate MRE and SDR in validation set
python main_baseline.py --test_flag 1 --resume ../SavePath/baseline/model.ckpt  # calculate MRE and SDR in test set

python main_yolol.py --test_flag 0 --resume ../SavePath/yolol/model.ckpt  # calculate MRE and SDR in validation set
python main_yolol.py --test_flag 1 --resume ../SavePath/yolol/model.ckpt  # calculate MRE and SDR in test set
```