Skip to main content

Typer CLI package for the Vit medical multi-axis Vision Transformer project.

Project description

btvit-cli

基于 PyTorch 的三轴 ViT 医学影像二分类 CLI。

安装

pip install btvit-cli
vit --version

当前主线

项目现在只保留一条流程:

  1. axis 阶段分别训练 x / y / z 三个轴的单轴 ViT
  2. fusion 阶段复用三轴 encoder,训练患者级 fusion_nn
  3. pred 阶段支持两种患者级输出:
    • weighted_vote
    • fusion_nn
    • both

每个患者的每张 (16, 368) 灰度图都作为独立样本共享患者标签,不再使用旧的中间特征缓存或跨轴组合穷举训练流程。

快速开始

1. 导出模板

vit get --path ./configs

导出内容:

  • model_config.yaml
  • data_config.yaml
  • training_config.yaml
  • experiment_config.yaml
  • multi_axis_config.yaml
  • monitoring_config.yaml
  • USAGE.md

2. 训练三轴 ViT

vit train --config configs/base --stage axis --axis x --verbose
vit train --config configs/base --stage axis --axis y --verbose
vit train --config configs/base --stage axis --axis z --verbose

关键参数:

  • --stage axis
  • --axis x|y|z
  • --config <yaml_dir>
  • --data 可选。未提供时使用 multi_axis.axis_data_paths.<axis>
  • --label 可选。未提供时使用配置文件中的标签路径

3. 训练 fusion 头

vit train --config configs/base --stage fusion --verbose

fusion 阶段会默认解析:

  • experiments/<name>/axis-x/checkpoints/best_model.pth
  • experiments/<name>/axis-y/checkpoints/best_model.pth
  • experiments/<name>/axis-z/checkpoints/best_model.pth

也可以显式覆盖:

  • --axis-checkpoint-x
  • --axis-checkpoint-y
  • --axis-checkpoint-z

4. 患者级预测

vit pred --mode weighted_vote --config configs/base --data dataset/predict_root
vit pred --mode fusion_nn --config configs/base --data dataset/predict_root
vit pred --mode both --config configs/base --data dataset/predict_root --label dataset/label.csv

预测模式:

  • weighted_vote:单图概率 -> 轴内患者均值 -> 跨轴加权投票
  • fusion_nn:单图特征 -> 轴内患者均值 -> fusion head
  • both:同时输出两条分支

预测数据支持:

  1. 根目录包含 data-x/data-y/data-z/
  2. 平铺 PNG 目录,文件名为 {patient_id}_{Z}_{Y}_{X}.png

输出结构

训练输出:

experiments/<experiment_name>/
  axis-x/
    checkpoints/
    results/
    logs/
  axis-y/
    checkpoints/
    results/
    logs/
  axis-z/
    checkpoints/
    results/
    logs/
  fusion/
    checkpoints/
    results/
    logs/
  testset/
    data-x/
    data-y/
    data-z/
    label.csv

预测输出:

  • predictions.csv
  • prediction_summary.json
  • confusion_matrix_weighted_vote.png,当有标签且启用 weighted_vote
  • confusion_matrix_fusion_nn.png,当有标签且启用 fusion_nn

配置要点

  • training.common:设备、worker、混合精度等公共训练参数
  • training.axis:单轴 ViT 训练参数
  • training.fusion:fusion 头训练参数
  • multi_axis.patient_pooling:患者内均值聚合规则
  • multi_axis.weighted_vote:加权投票配置
  • multi_axis.fusion_head:患者级 fusion 头配置
  • monitoring.axis / monitoring.fusion / monitoring.prediction:分阶段监控与预测阈值

开发

conda activate Vit
pip install -e .
vit --help

构建:

conda run -n Vit python -m build

许可证

Apache-2.0

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

btvit_cli-0.1.9.tar.gz (132.6 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

btvit_cli-0.1.9-py3-none-any.whl (121.8 kB view details)

Uploaded Python 3

File details

Details for the file btvit_cli-0.1.9.tar.gz.

File metadata

  • Download URL: btvit_cli-0.1.9.tar.gz
  • Upload date:
  • Size: 132.6 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.12.12

File hashes

Hashes for btvit_cli-0.1.9.tar.gz
Algorithm Hash digest
SHA256 77fd837a00210dc09766a02388553613136683bd5a6b04585f0a8e8e1955edd1
MD5 c462276e733aadf7b7b2c61ae2f7c7d7
BLAKE2b-256 e4f28e2c888b3cdbbaa622196625a563c9dd5b8bc27b2b96cf98814e409320d4

See more details on using hashes here.

File details

Details for the file btvit_cli-0.1.9-py3-none-any.whl.

File metadata

  • Download URL: btvit_cli-0.1.9-py3-none-any.whl
  • Upload date:
  • Size: 121.8 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.12.12

File hashes

Hashes for btvit_cli-0.1.9-py3-none-any.whl
Algorithm Hash digest
SHA256 d4fd0abd12df30a5c71836e84d3ac7770f7897229bc18f12c77f4049d5e0cd53
MD5 fefe27b3351b0a8c7d3593e28d474de1
BLAKE2b-256 f62a645d085d632c796a7c3094394611412ee5a095be21385e41e1f2afc4cdf8

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page