Plant Pathology 2020 - FGVC7

About

My solution to the Plant Pathology 2020 - FGVC7 competition on Kaggle.

As it was a late submission, I tried to improve on the first place solution.

Score

• Private Score: 0.96834
• Public Score: 0.96732

Usage

Install dependencies

poetry install
poetry run wandb login

You can also use DevContainer.

Execute

Step 1: Train the model using k-fold cross validation(k=5).

python train.py --train_batch_size 32 --gpus 0

Step 2: Generate soft labels for self-distillation training.

python generate_soft_labels.py

Step 3: Use soft and hard labels and train the model using k-fold cross validation(k=5).

python train.py --train_batch_size 32 --gpus 0 --soft_labels_filename soft_labels.csv --log_dir logs_submit_distill

Step 4: Generate the results of the model predictions generated by the distillation.

python generate_distill_submission.py

Step 5: Generate final results

python generate_final_submission.py

Contributing

Your contribution is always welcome. Please read Contributing Guide.