kaggle_bengali_recognition

model summary excel and model weights: https://oworkspace-my.sharepoint.com/personal/tonghu_oocl_com/_layouts/15/onedrive.aspx?viewid=a1352ee7-5566-48ae-8275-879ac9c7978d&id=%2Fpersonal%2Ftonghu_oocl_com%2FDocuments%2Fkaggle%2Fbengali

Pre-process data

1. Run data_processing_final.ipynb (Run until the Step by Step cell)
2. Run for from train_image_data_0.parquet to train_image_data_3.parquet
3. Save the pre-processed pickle data into data/processed_data/size_224/

Run training

Suggest running model_3_efficient_net.ipynb

Run inference

In model_weights/eff_0_baseline or model_weights/eff_0_with_mixup_cutmix, there are 5 folds of trained model weights.

Run kaggle_submission.ipynb for inference.