project-monai / monai Goto Github PK

View Code? Open in Web Editor NEW

5.7K 89.0 1.0K 67.79 MB

AI Toolkit for Healthcare Imaging

Home Page: https://monai.io/

License: Apache License 2.0

Python 95.01% Shell 0.23% Dockerfile 0.03% C++ 2.49% Cuda 2.08% C 0.16%

healthcare-imaging deep-learning medical-image-computing medical-image-processing pytorch python3 monai

monai's Issues

Torchscript-safe variable layer count module

Create hard-coded reference networks for each canonical paper implementation network that we include in MONAI

model loader/saver with pytorch state_dict

58 preprocessor adaptor function for compose

Implement sequential Compose functionality

setup a documentation website auto-generated from code docstrings

Additional comment from #21

We should also nail down what format comments should be in, I would suggest Numpy format https://numpydoc.readthedocs.io/en/latest/format.html as a good hybrid of Google docstrings and reStructured Text formats, however Pytorch uses Google docstrings so maybe we should stick to that.

Create a GitHub PR and issue template

Port loss functions from niftynet and clara train

intensity normalisation transform

Implement Transforms and test Pre-processing pipeline. Identify the inventory of transforms from Clara, NiftyNet and Eric, Ben's archive

Port dense v-net

a subtask of #78, port dense v-net

ref:
https://niftynet.readthedocs.io/en/dev/niftynet.network.dense_vnet.html

@eligibson

Port metrics from niftynet and clara train

Update preprocessing design page on the wiki

trainer based on components and ignite

Simplified integration tests with an end-to-end example

initial contribution guidelines documentation

MVP example for model inference and saving predicted results.

Sliding window inference
Writing segmentation as a nifti image (Nifti image writer)
Writing classification as a json file (json writer)

Create layer factory mechanism

model evaluator based on components and ignite (computing validation scores from training model)

coding style

(This ticket is originally from the monai-minimal repo)

Would be great to have a good coding style from day one...
e.g.,
Function and variable names should be lowercase, with words separated by underscores as necessary to improve readability.
Class names should normally use the CapWords convention.

To close this ticket, we ensure that on the project running flake8 (with pep8-naming plugin) with config https://github.com/Project-MONAI/MONAI/blob/a106c79b706a72933520aaba4fc43eb85765985e/.flake8
returns no error message.

2D/3D UNet architecture

Documentation Content Completeness

Create library of standard convolutional blocks, such as residual units

Finalize Folder Structure

create CI/CD pipelines, so that the project has automated code checks for pull requests

Finalize the design pattern for Transform.

code examples of MONAI input data pipeline

implement reference networks from papers using pure pytorch

training stats event-handler (for printing training losses)

sliding window generator

Tutorials & Get Started Guides for Public Alpha

MVP example for model validation every N training steps

-Depends on task 2

Minimal evaluation metrics interface (or using ignite.metrics?)
-Event handler for model validation and early stopping

Design Discussions

Definition of Networks, standardized way of defining network parameters

Prepare the codebase for the public alpha:

-Refactor input ‘transform’ interfaces
- Finalise the code folder structure
- More thorough unit tests
- More integration tests
- Following the current design, more networks/losses/metrics/transforms

MVP example for multiple GPU training

-Additional refactoring iteration to unify single-gpu/multiple gpu training?

random window cropping transform

Design Discussions for Model Serialization & De-Serialization.

Standardize Data Loader for Challenges Data (Decath data)

e.g. Unzip .tar; read .json
definition and create a standardized way of constructing input data pipelines

Port layers across from ClaraTrain

Port highresnet

a subtask of #78, port highresnet

ref:
https://niftynet.readthedocs.io/en/dev/niftynet.network.highres3dnet.html
https://github.com/fepegar/highresnet

project-monai / monai Goto Github PK

monai's Issues

Recommend Projects

Recommend Topics

Recommend Org