Kaggle Clouds Segmentation Challenge

• Kaggle Clouds Segmentation Challenge
  • Requirements
  • Features
  • Folder Structure
  • Usage
  • Components
    • Model
    • Loss
    • metrics
    • Validation data
  • Tensorboard Visualization
  • TODOs
  • License
  • Acknowledgements

Requirements

• Python >= 3.5 (3.6 recommended)
• PyTorch >= 1.1 (1.2 recommended)
• tqdm (Optional for test.py)
• tensorboard >= 1.14 (see Tensorboard Visualization)
• torchvision
• numpy
• tqdm
• tensorboard>=1.14
• catalyst
• matplotlib
• albumentations
• segmentation-models-pytorch
• seaborn
• scikit-learn

Folder Structure

kaggle-clouds-segmentation-challenge/
│
├── train.py - main script to start training
├── test.py - evaluation of trained model
├── data_loader/ - anything about data loading goes here
│   └── data_loaders.py
|
├──saved_models/ - contains the saved model weights after the training
├── data/ - default directory for storing input data
│
├── model/ - models, losses, and metrics
│   ├── model.py
|
├── saved/
│   ├── models/ - trained models are saved here
│  
│
└── utils/ - small utility functions
    ├── utils.py

Usage

Download the dataset from kaggle:

• Install kaggle using pip

pip install kaggle

• Go to your kaggle account and download the API token and place it in the ~/.kaggle/ directory
• Download the dataset:

kaggle competitions download -c understanding_cloud_organization

• unzip the dataset into our project folder

cd kaggle-clouds-segmentation-challenge
mkdir dataset

# Go to the folder where the dataset is downloaded and use the below command
unzip understanding_cloud_organization.zip -d kaggle-clouds-segmentation-challenge/dataset/

python train.py -bs <batch_size:int> -epochs <num_epochs:int>

Default values:
* batch_size : 16
* num_epochs: 20

Testing using the saved model:

• This generates a submission file for kaggle Provide paths to the saved model after training and path to the submission file downloaded from kaggle in the first step for the below command

python3 test.py --model_path <path/to/saved_model/> --sub_path <path/to/submission/file> 

Tensorboard Visualization

Visualization of Experiments using TensorBoard

Link to TensorBoard

TODOs

• Resume Checkpoints
• Implement Callbacks
• Automate writing to submission files
• Enable TensorBoard Logging

License

This project is licensed under the MIT License. See LICENSE for more details

Acknowledgements