/Reproduce_CutMix

Reproduce CutMix algorithm with the paper "CutMix: Regularization Strategy to Train Strong Classifiers with Localizable Features"

Primary LanguageJupyter Notebook

Reproduce CutMix algorithm

In this repository,

We are going to reproduce the CutMix algorithm based on the paper "CutMix: Regularization Strategy to Train Strong Classifiers with Localizable Features". ---The project was finished on 2022/05/24

The Overall Process

  1. Introduce problem
  2. Motivations
  3. Datasets used
  4. Main methods
  5. tweaks and evaluation

+a ) we will add extra codes to visualise additional details that paper may not have shown. Above tasks will be done with the code demo.

Check list

  1. data loading [22.5.6]
  2. methods (including NNs) [22.5.8]
  3. model compilation [22.5.8]
  4. loss functions and optimisers [22.5.8]
  5. training [22.5.8-22.5.24]
  6. tweaks [o]
  7. comparison cutmix regulariztion with others [o]
  8. feature extraction vs fine-tunning [o]

  1. Other dataset[BegaliDB]

Development Environment & Tools

  1. Colab
  2. Jupyter notebook
  3. Visual Studio Code
  4. Github
  5. Python
  6. Pytorch

Schedule

[22.5.6] Initiate the project

[22.5.6] Finishing Environment Setup

[22.5.6] Data download

[22.5.8] Training

[22.5.20] paper summary

[22.5.2] presentation

-----Project Finished-----

Reference

paper link ↓↓↓↓↓↓↓↓

ref) https://arxiv.org/abs/1905.04899v2

github link ↓↓↓↓↓↓↓↓

main ) https://github.com/clovaai/CutMix-PyTorch

sub ) https://github.com/kaggler-tv/codes

Dataset link

kaggle ) https://www.kaggle.com/c/bengaliai-cv19/data

GIT tutorial

git init

git add .

git commit -m "어디어디 수정"

git remote add origin https://github.com/본인아이디/레포지토리이름.git

git push -u origin master