• 주최 : LG AI Research
• 주관 : DACON
• https://dacon.io/competitions/official/235870/overview/description

/workspace
├── data
│   ├── train
│   │    ├── 10027
│   │         ├── 10027.csv
│   │         ├── 10027.jpg
│   │         └── 10027.json
│   │    ├── ...
│   │    └── 67678
│   ├── test
│   │    ├── 10000
│   │    ├── ...
│   │    └── 67677
│   │    
│   ├── train.csv
│   └── sample_submission.csv
│
├── main.ipynb (실행 코드 - jupyter notebook)
├── baseline.py (실행 코드)
├── dataset.py (데이터셋 클래스)
├── model.py (모델 클래스)
├── loss.py (손실함수 클래스)
├── single_gpu_inference.py (단일 GPU로 추론하는 코드)
├── image_model_list.txt (참고 : 사용 가능한 이미지 모델 이름)
├── requirement.txt
├── Dockerfile   
└── docker-compose.yml

1. Install Library

   pip3 install -r requirement.txt
   pip3 install jupyter
   

2. Download data.zip from https://dacon.io/competitions/official/235870/data to container workspace data path.

   #./LG_Plant_Disease_Diagnosis
   mkdir data
   cd data
   (Download data to ./LG_Plant_Disease_Diagnosis/data/)

3. Unzip train, test data

   #./LG_Plant_Disease_Diagnosis/data
   unzip data.zip
   unzip train.zip
   unzip test.zip

4. Train main.ipynb

5. Submit ./submission_xxx.csv

1. git clone https://github.com/glee1228/LG_Plant_Disease_Diagnosis.git

2. Edit docker-compose.yml

   services:
     main:
       container_name: plant-lg-dacon
       build:
           context: ./
           dockerfile: Dockerfile {If Ubuntu version is 20.04, Edit it w/ Dockerfile2}
       ...
       ports:
         - "{host ssh}:22"
       ipc: host
       stdin_open: true
   

3. Download data.zip from https://dacon.io/competitions/official/235870/data to container workspace data path.

   #./LG_Plant_Disease_Diagnosis
   mkdir data
   cd data
   (Download data to ./LG_Plant_Disease_Diagnosis/data/)

4. Build docker image clearly and create containers

   #./LG_Plant_Disease_Diagnosis
   docker-compose build --no-cache
   docker-compose up -d
   docker attach plant-lg-dacon

5. Unzip train, test data

   #/workspace/data
   unzip data.zip
   unzip train.zip
   unzip test.zip

6. (Option) Set password and Restart SSH for SFTP connection

   passwd
   /etc/init.d/ssh restart

7. Train baseline.py

   #/workspace
   python baseline.py

8. Submit /workspace/submission_xxx.csv

1. edit single_gpu_inference.py

Enter the paths of 5 models as a string in the model_path_list list.(line192)

``` 
model_path_list = [model path 1..,
                   model path 2..,
                   model path 3..,
                   model path 4..,
                   model path 5..]

```

2. inference using single GPU

   #/workspace 
   python single_gpu_inference.py

Ubuntu 18.04.5 LTS

• h5py>=2.10.0
• numpy>=1.18.1
• tqdm>=4.43.0
• albumentations==1.1.0
• matplotlib==3.5.1
• opencv-python-headless==4.5.5.62
• pandas==1.3.5
• Pillow==9.0.0
• scikit-image==0.19.1
• scikit-learn==1.0.2
• scipy==1.7.3
• timm==0.5.4
• torch==1.8.0
• torch-optimizer==0.3.0
• torchvision==0.9.0
• wandb==0.12.9
• easydict==1.9