DeepCenterline
所属分类：数值算法/人工智能
开发工具：TeX
文件大小：602230KB
下载次数：0
上传日期：2022-05-02 22:37:52
上 传 者：
sh-1993
说明： DeepCenterline，用于找到血管分割的中心线距离图的模型，然后将其输入到图中的最小路径ex...
(Model for finding a centerline distance map of blood vessel segmentations to then be fed into a graph minimum path extractor to find the centerlines of a vessel using PyTorch. It uses a Residual UNet (Convolutional AutoEncoder) type architecture.)
文件列表:
.vscode (0, 2022-05-03)
.vscode\settings.json (279, 2022-05-03)
Centerline Extraction Project - Kyle Beggs.pptx (23257420, 2022-05-03)
Project ideas.pdf (505578, 2022-05-03)
data (0, 2022-05-03)
data\DeepVesselNet (0, 2022-05-03)
data\DeepVesselNet\centerline (0, 2022-05-03)
data\DeepVesselNet\centerline\1.nii.gz (1142047, 2022-05-03)
data\DeepVesselNet\centerline\10.nii.gz (1146812, 2022-05-03)
data\DeepVesselNet\centerline\100.nii.gz (1146520, 2022-05-03)
data\DeepVesselNet\centerline\101.nii.gz (1148426, 2022-05-03)
data\DeepVesselNet\centerline\102.nii.gz (1148823, 2022-05-03)
data\DeepVesselNet\centerline\103.nii.gz (1140302, 2022-05-03)
data\DeepVesselNet\centerline\104.nii.gz (1144662, 2022-05-03)
data\DeepVesselNet\centerline\105.nii.gz (1141694, 2022-05-03)
data\DeepVesselNet\centerline\106.nii.gz (1148812, 2022-05-03)
data\DeepVesselNet\centerline\107.nii.gz (1151033, 2022-05-03)
data\DeepVesselNet\centerline\108.nii.gz (1143565, 2022-05-03)
data\DeepVesselNet\centerline\109.nii.gz (1140515, 2022-05-03)
data\DeepVesselNet\centerline\11.nii.gz (1144638, 2022-05-03)
data\DeepVesselNet\centerline\110.nii.gz (1145746, 2022-05-03)
data\DeepVesselNet\centerline\111.nii.gz (1142568, 2022-05-03)
data\DeepVesselNet\centerline\112.nii.gz (1146579, 2022-05-03)
data\DeepVesselNet\centerline\113.nii.gz (1141243, 2022-05-03)
data\DeepVesselNet\centerline\114.nii.gz (1145512, 2022-05-03)
data\DeepVesselNet\centerline\115.nii.gz (1142952, 2022-05-03)
data\DeepVesselNet\centerline\116.nii.gz (1142523, 2022-05-03)
data\DeepVesselNet\centerline\117.nii.gz (1133380, 2022-05-03)
data\DeepVesselNet\centerline\118.nii.gz (1139965, 2022-05-03)
data\DeepVesselNet\centerline\119.nii.gz (1143420, 2022-05-03)
data\DeepVesselNet\centerline\12.nii.gz (1145722, 2022-05-03)
data\DeepVesselNet\centerline\120.nii.gz (1142162, 2022-05-03)
data\DeepVesselNet\centerline\121.nii.gz (1139581, 2022-05-03)
data\DeepVesselNet\centerline\122.nii.gz (1136060, 2022-05-03)
data\DeepVesselNet\centerline\123.nii.gz (1144568, 2022-05-03)
data\DeepVesselNet\centerline\124.nii.gz (1142686, 2022-05-03)
data\DeepVesselNet\centerline\125.nii.gz (1140493, 2022-05-03)
... ...
# Centerline Extraction
Model for finding a centerline distance map of blood vessel segmentations to then be fed into a graph minimum path extractor to find the centerlines of a vessel using PyTorch. It uses a Residual UNet (Convolutional AutoEncoder) type architecture.
The [train.py](https://github.com/kylebeggs/DeepCenterline/blob/master/src/train.py) file in the src directory and the models are defined in [models.py](https://github.com/kylebeggs/DeepCenterline/blob/master/src/models.py). There is a helper file with some plotting functions,
etc named [utils.py](https://github.com/kylebeggs/DeepCenterline/blob/master/src/utils.py).
We make use of argparse here, so simply type ``` python train.py --help ``` to see the potential training settings as such:
```shell
> python train.py --help
usage: train.py [-h] --name NAME [--epochs EPOCHS] [--lr LR] --model MODEL
optional arguments:
-h, --help show this help message and exit
--name NAME Name the model you are training.
--epochs EPOCHS Set max number of epochs.
--lr LR, --learning-rate LR
Set the initial learning rate to be used with the reduced step scheduler.
--model MODEL Set which model to use. Options are pooling or striding.
```
Providing a name is required. ```--epochs``` default is 100 and ```--lr``` default is 1e-3. ```--model``` sets which approach to use, striding or pooling.
The file outputs a TensorBoard file in the [runs/](https://github.com/kylebeggs/DeepCenterline/blob/master/runs) directory with the name which you gave as input as seen above. Launch the TensorBoard session with ```tensorboard --logdir=runs``` from the main directory of this repo. A log file is output containing the training settings and test set performance metrics such as accuracy and F1 score in the [logs/](https://github.com/kylebeggs/DeepCenterline/blob/master/logs) directory. The model weights are saved on the last epoch into the [trained_models/](https://github.com/kylebeggs/DeepCenterline/blob/master/trained_models) directory named according to the name you provide when launching the training. The train file automatically detects if you have a CUDA enabled device and selects it for training with a fallback to the CPU.
[Here is the report for the course this code was developed for.](https://github.com/kylebeggs/DeepCenterline/blob/master/report/kyle_beggs_report.pdf)
近期下载者：
相关文件：
收藏者：