Korea landmark recognition

Development of Artificial Intelligence Algorithm to Classification Landmark Images from Large-Scale Images [link].

Extract korea landmark data

Please download and extract image data in the public

$ cd landmark
$ mkdir public
$ wget https://dacon-datasets.s3.ap-northeast-2.amazonaws.com/landmark/data.zip
$ unzip data.zip -d public

Input data

└── landmark
    └── public
        └── train
            └── 부산시
                └── 사직야구장
                    └── xxx.JPG (landmark name)
        └── test
            └── x (name doesn't matter)
                └── xxx.JPG (landmark name)
        └── category.csv
        └── category.json
        └── train.csv
        └── sample_submission.csv

Usage

Training

Run commands as follows:

$ cd landmark
$ python main.py --gpu 0 --epochs 100 --batch_size 80 --image_size 224 --model_dir save/test_01 --depth 0

Testing

Run commands as follows:

$ cd landmark
$ python main.py --gpu 0 --load_epoch 99 --batch_size 80 --image_size 224 --test --model_dir save/test_01 --test_csv_submission_dir my_submission.csv

Requirements

• python 3.6
• torch 1.4
• albumentations
• torch_optimizer
• efficientnet_pytorch
• matplotlib
• numpy
• pandas
• sklearn
• tqdm
• argparse