Skip to content

69th(Silver) Solution of "Understanding Clouds from Satellite Images | Kaggle" with PyTorch-Lightning

Notifications You must be signed in to change notification settings

yukkyo/Kaggle-Understanding-Clouds-69th-solution

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

34 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Kaggle-Understanding-Clouds-69th-solution

https://www.kaggle.com/c/understanding_cloud_organization/overview

0. Environment

  • Python 3.6.5
  • CUDA 10.1.243

1. Preparation

1.1 install Python packages

$ pip install -r docker/requirements.txt

1.2 Download dataset

Use kaggle API command.

input
├── sample_submission.csv
├── test_images
├── train.csv
└── train_images

1.3 Split kfolds

$ cd src
$ python data_process/s01_make_kfold_csv.py --kfold 5

2. How to train

$ cd src
$ python train.py --config configs/model063.yaml --kfold 1
$ python train.py --config configs/model063.yaml --kfold 2
$ python train.py --config configs/model063.yaml --kfold 3
$ python train.py --config configs/model063.yaml --kfold 4
$ python train.py --config configs/model063.yaml --kfold 5
$ python train.py --config configs/model064.yaml --kfold 1
$ python train.py --config configs/model064.yaml --kfold 2
$ python train.py --config configs/model064.yaml --kfold 3
$ python train.py --config configs/model064.yaml --kfold 4
$ python train.py --config configs/model064.yaml --kfold 5

3. How to predict test

$ cd src
$ python submission.py --kfolds 12345 --config configs/model063.yaml --use-best --save-predicts
$ python submission.py --kfolds 12345 --config configs/model064.yaml --use-best --save-predicts
$ python utils/make_mean_sub.py
$ ls ../output/final
sub_model063064_kfolds12345_top062_minarea16250-13750-10000-5000_bottom042_usebest_6364avg_thres.csv

About

69th(Silver) Solution of "Understanding Clouds from Satellite Images | Kaggle" with PyTorch-Lightning

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published