DLMarines

Team: Bartosz Brzoza, Magdalena Buszka, Martyna Firgolska, Michał Kulibiński

Description: This project contains simple workflow for classification of images of marine animals, leveraging kedro framework. It was made as a student project for "Projeect: Deep Learning" course at University of Wrocław.

Dataset

The dataset contains images of marine animals - 23 different classes (Seahorse, Nudibranchs, Sea Urchins, Octopus, Puffers, Rays, Whales, Eels, Crabs, Squid, Corals, Dolphins, Seal, Penguin, Starfish, Lobster, Jelly Fish, Sea Otter, Fish, Shrimp and Clams). Each image size is of the type (k, 300px) or (300px, k), where k is a number less or equal to 300. Example images:

How to use:

Begin by downloading the repository by cloning it or downloading the zip to the directory of your choice and openig the repo's folder.

Installation:

To install the environment run the commands below

conda env create  --file conda.yml
conda activate dlmarines
poetry install

Data downloading

You can download data manually from https://www.kaggle.com/datasets/vencerlanz09/sea-animals-image-dataste into data/01_raw or use data_downloading pipeline by running

kedro run --pipeline=data_downloading

Note that the pipeline uses kaggle api, so in order to run it follow the steps below to download your kaggle key.

Read more about data downloading pipeline.

Download Kaggle Api Key:

1. Sign in to kaggle
2. Go to Account
3. Go to API section and click Create New API Token. It will download kaggle.json with your username and key.

{ "username":"your_kaggle_username","key":"123456789"}

4. In conf/local/credentials.yml add your username and key as shown below:

kaggle:
      username: "your_kaggle_username"
      key: "123456789"

Data preprocesing

To preprocess data from /data/01_raw/sea-animals-image-dataste.zip use data_processing pipeline

kedro run --pipeline=data_processing

Read more about data preprocessing pipeline.

Model training

To train the model use model_training pipeline

kedro run --pipeline=model_training

Read more about model training pipeline.

Model evaluation

To evaluate the model use model_evaluation pipeline

kedro run --pipeline=model_evaluation

Read more about model evaluation pipeline.

Running:

To run all pipelines you can use command:

kedro run

Remember that in order to automatically download dataset you need to add your kaggle key.

About the project

Weights&Biases

Here you can view the Weights&Biases report

Documentation

Detailed documentation can be found here

Libraries and technologies

The technologies and main libraries used in the project:

• Kedro - for project framework and structure
• mkdocs - for documentation
• Poetry - for managing dependencies
• PyTorch
• Lightning - for easy implementation of models and training