https://github.com/mdciri/teeth-segmentation
teeth segmentation using pytorch and monai
https://github.com/mdciri/teeth-segmentation
deep-learning gui image-segmentation monai pytorch teeth-segmentation unet-image-segmentation
Last synced: 2 months ago
JSON representation
teeth segmentation using pytorch and monai
- Host: GitHub
- URL: https://github.com/mdciri/teeth-segmentation
- Owner: mdciri
- Created: 2022-06-26T11:38:04.000Z (about 3 years ago)
- Default Branch: main
- Last Pushed: 2023-03-23T22:24:05.000Z (over 2 years ago)
- Last Synced: 2023-06-05T02:15:33.967Z (about 2 years ago)
- Topics: deep-learning, gui, image-segmentation, monai, pytorch, teeth-segmentation, unet-image-segmentation
- Language: Python
- Homepage:
- Size: 813 KB
- Stars: 4
- Watchers: 1
- Forks: 1
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Teeth Segmentation
Teeth segmentation using Pytorch and [Monai](https://monai.io/). The data used belong to the [Tufts Dental Database](http://tdd.ece.tufts.edu/).
# Data split
The Thfts dataset is composed by 1000 images in total which were randomly divided as 70-20-10% in training, validation, and test set. The split is done by:
pyhton data_split.py
which save the dataset lists in the `data.json` file. It stores also the class names and the class weights for the segmentation.
# Model
Simple UNet network created using `monai.netowrks`.
# Training
pthon train.py -md "" -d "mps" -g 0 -bs 16 -lr 1.e-4 -ne 100
This scripts runs the training scrits. It:
- saves the best model at the specified `""`,
- uses the device `"mps"` (but you can use `"cuda"` or `"cpu"`)
- GPU at position `0`,
- batch size `16`,
- learning rate `1.e-4`,
- `100` number of maximum epochs.
# Evaluation
pthon evaluation.py -md "" -d "mps" -g 0 -bs 1
This script computes evaluation over the training, validation, and test dataset saving the dice score per each image in the file `evaluation_results.csv`.
# GUI
pthon gui.py
This scripts runs a simple GUI where the user can upload an image, visualize the model's segmentation prediction, and, in case, save it.
