Objective

The main purpose of the project is to improve ML in Production skills, research how image classification models work.

Data

The data we are using are taken from Planet: Understanding the Amazon from Space Kaggle competition. More info can be found at the competition website. In a nutshell we want to label satellite image chips with atmospheric conditions and various classes of land cover/land use. It is a multi-labeling problem with 17 different classes. In the competition algorithms were scored using the mean F2 score.

Here we only use the jpg images. Note that zip file should be unzipped.The data can still be downloaded here

Pipeline preparation

1. Creating and activating the environment

   python3 -m venv /path/to/new/virtual/environment
   

   source /path/to/new/virtual/environment/bin/activate
   

2. Installing packages

   In the activated environment:

   pip install -r requirements.txt
   

3. Customise config.yaml to suit your needs. Pay attention to config.train_classes_path, you need to specify where the dataset file train_classes.csv was downloaded to, and in data_dir you need to specify where the train-jpg folder was downloaded to.

Training

Start training:

PYTHONPATH=. python src/train.py configs/config.yaml

Load best model using dvc

Command

dvc pull weights/vgg16_feature_extractor.pth.dvc

will create vgg16_feature_extractor.pth within weights directory, it will be used as a base model for prediction

Inference

command

PYTHONPATH=. python src/predict.py --image_path <path_to_image>

will be used as a inference point for <path_to_image> image, test_image is provided in the root of repo

Experiment Logs

Link to VGG16 experiment in ClearML