README.md

Clouds Satellite

Classify cloud structures from satellites.

How to run:

bash sh scripts/train.sh
bash sh scripts/test.sh
bash sh scripts/submit.sh

Tensorboard:

tensorboard --logdir=configs/`config_folder`/log/

Docker

make build
make run
make exec

Configs file structure

├── configs
│   ├── config_folder
│   │   |
│   │   │── config_name.yaml
│   │   │── inference_config_name.yaml
│   │   │── submit_config_name.yaml
│   │   ├── checkpoints
│   │   │      ├── foldi
│   │   │      │   ├── topk_checkpoint_from_fold_i_epoch_k.pth 
│   │   │      ├── best
│   │   │      |   ├── best_checkpoint_foldi.pth
│   │   ├── log
|   |   |      ├── logs_fold_i.txt
|   |   ├── dict_efficientnet-b3
|   |   |       ├── experiment_foldi_epochi.pkl
|   |   |   
├── cloudsimg
|   |   |
│   ├── test_images
|   |   | ── *.jpg
|   |   | ── *.jpg
|   ├── train_images
|   |   | ── *.jpg
|   |   | ── *.jpg
|   ├── sample_submission.csv
|   ├── train.csv

Install

pip install -r requirements.txt

References

• https://github.com/albu/albumentations
• https://github.com/qubvel/segmentation_models.pytorch
• https://github.com/amirassov/kaggle-pneumothorax
• https://github.com/sneddy/pneumothorax-segmentation