rosklyar/no_waste_com_ua

nowaste.com.ua secondary materials classifier project

Project description

This project is a part of the nowaste.com.ua project. It is a classifier for secondary materials.

How to train models

Download data

1. Download data from Kaggle Garbage Classification. Or use your own dataset with the same structure(classes are separeted by folders).
2. Unpack data to ./data/kaggle-ds folder for example.

Training routine

You can train model using next cli commands consecutively:

1. Separate data to train and test sets

python .\src\garbage_classifier\cli.py split-data-folder .\data\kaggle-ds .\data\splitted\ 0.2

Output: .\data\splitted\train and .\data\splitted\test folders with train and test data. 2. Train model Login to wandb using wandb login command. Then run training:

python .\src\garbage_classifier\cli.py train-and-log-wandb .\config\training.json .\data\splitted\train .\data\splitted\test .\data\model\output  special-project

Or you can run training without logging to wandb:

python .\src\garbage_classifier\cli.py train-and-save .\config\training.json .\data\splitted\train .\data\splitted\test .\data\model\output

Output: model and it's model card inside .\data\model\output folder. 3. Upload model to wandb registry Note! You need to login to wandb using wandb login command.

python .\src\garbage_classifier\cli.py upload-to-registry wandb-entity-name special-project model-name .\data\model\output .\config\classes.json