In this project, I provided an simple end-to-end implemetation for training, evaluating and monitoring a stardard classifier which is considered as a common template and easy-to-use for AI newbies. I did apply several techniques such as Expotential Moving Average (EMA), Label Smoothing, Modern Optimizer, K-Fold cross-validation, random-augmentations ... traking logs, hyperparameters ... with comet ML tool.
comet_ml
timm
opencv-python
torch
torchvision
Here we want to test our method with 5-Fold Cross-Validation. We just need to put FOLD_i into folder ./data.
First, set the value of ROOT in config file config.yml so that you can train the case(Fold_i) you want.
Then, for experimental logging info, I used framework comet_ml. We need to create a file called experiment_apikey.txt. This file will just contain your api_key that the main website of comet_ml provides you when you create your own account.
For create data form for dataset, we need config files. Here just run python3 utils.py
For training, we run python3 main.py