kaggle-understanding-clouds

Code for 1st place solution in Kaggle Understanding Clouds from Satellite Images Challenge.

To read the brief description of the solution, please, refer to the Kaggle post

Reproducing Submission

To reproduce my submission without retraining, do the following steps:

1. Installation
2. Download Dataset
3. Download Pretrained models
4. run bash reproduce.sh

Installation

All requirements should be detailed in requirements.txt. Using Anaconda is strongly recommended.

conda create -n cloud python=3.6
conda activate cloud
pip install -r requirements.txt

Prepare dataset

Download dataset

Download and extract train_images.zip and test_images.zip to data directory.

$ kaggle competitions download -c understanding_cloud_organization
$ unzip understanding_cloud_organization.zip -d data
$ chmod 644 data/*
$ unzip data/train_images.zip -d data/train_images
$ unzip data/test_images.zip -d data/test_images

Generate CSV files

$ python tools/split.py

Resize images and labels

$ python tools/resize_images.py
$ python tools/resize_labels.py

Training

In the configs directory, you can find configurations I used to train my final models.

Train models

To train models, run following commands.

$ python run.py train with {config_path} -f

Average weights

To average weights, run following commands.

$ python run.py swa with {config_path} swa.num_checkpoint=5 swa.ema=0.33 swa.epoch_end=40 -f

The result will be located in train_logs/{train_dir}/checkpoint.

Pretrained models

You can download pretrained model that used for my submission from link

$ mkdir checkpoints
$ bash download_pretrained.sh

Inference

If trained weights are prepared, you can create files that contains class/mask probabilities of images.

$ python run.py inference with {config_path} \
  inference.output_path={output_path}  \
  transform.params.tta={1..4} \
  inference.split={split} \
  checkpoint={checkpoint_path}

Evaluate

To evaluate dev/test_dev set, run following commands.

python tools/evaluate --input_dir {comma seperated list of inference_result_paths}

Make Submission

python tools/make_submission.py \
  --input_dir {comma seperated list of inference_result_paths} \
  --output {output_path}