Auther: Shizhe Cai Maastricht University
This whole repo will be reform
This repo was originally used as a tutorial material on how to use fork and git clone. However, although many geeks may simply delete this repe due to its uselessness, I decided to make it become my trick bag on my way learning programming (python).
| Folder name | ref |
|---|---|
| CommandLines | link |
| LearnDeepReg | link |
| LearnDocker | link |
| LearnGeneralPython | -- |
| LearnHF | link |
| LearnLightning | link |
| LearnOpen3D | link |
| LearnRAG | -- |
| LearnTorchIO | link |
T: a set of tutorials to learn, including text or video
I: most important information
I(t): a subset of I, most information from tutorial t
P: potential project to contribute
For t in T:
1. briefly check all content of t (~5 mins), decide:
if t is usefull for P:
2. find the most important part I(t) to learn, decide:
if I is hard to understand:
3. practice N times until understand I(t)
else:
continue
4. take notes on I(t) as reminder
else:
continue next t
For I(t) in I:
if I(t) directly related to P:
1. analyze functionality of I(t)
2. merge into P
3. think potential use-case of I(t)
K: the code for research
RQ: research question
E: experiments that validate the research
0. create a new repo and make a very good README file
1. design a RQ, and its related E
2. build up bedrock code K for baseline RQ and E
3. define the innovation section IS from K
4. change only the IS with same input and output
5. run multiple experiments to valid the RQ
Learning goal:
- learn the official tutorials from Lightning docs,
- learn the difference between pytorch training and lgihtning, whaty is pros and cons
- learn how to use lightning existed recipes to build up training framework for tasks like classification, segmentation, object detection ...
- learn how to customize the lightning existed framework elements, like loss function, optimizer, models, and metrics (with other packages)
- learn how to build a small project that can segment 3D medical images by using config to control different styles (medical image segmentation)
- use basic augmentation methods, affine, elasitc
- use at least three models with fine-tuning ability
- use metrics that can valid the results easily (use wandb)
- identify the switchable part of this project, dataset, loss, augmentation, models, optimizer and optimization methods
- build a inference function as a finish touch
Folder: LearnLightning
Details are included in ./LearnLightning/README.md