5th place solution of "AI× 商標：イメージサーチコンペティション（類似商標画像の検出）"

Description

Nanika Kaku

Setup

COMPETITION_NAME='nishika-22'

git clone https://github.com/Ynakatsuka/$COMPETITION_NAME
cd $COMPETITION_NAME

# Credentials
cp .env.template .env
# vim .env  # Set your credentials

# Download data & unzip to `data/input/`

# Delete empty directories
find data/input/cite_images/ -empty -type d -delete

If you don't use docker, you need to add PYTHONPATH

export PYTHONPATH="${PWD}/src:$PYTHONPATH"

How to Run

• Preprocess and train models and inference

docker-compose up -d
docker-compose exec kaggle ./bin/final.sh

• Benchmark test

docker-compose exec kaggle python src/benchmark/benchmark.py

• Demo App

docker-compose exec kaggle streamlit run src/benchmark/app.py

Then, access here: http://localhost:8501/

Others

• Format code

pysen run lint
pysen run format

References

• lightning-hydra-template
• kvt