Data-Centric Land Cover Classification

Description

This project focuses on developing an AI-driven ranking system to automatically assess and rank images based on their levels of label noise in a semantic segmentation dataset. The goal is to create a ranking that aligns with an undisclosed ground-truth ranking, evaluated using the Kendall Tau score. By effectively identifying noisy samples before training, this approach enhances data reliability, ultimately improving model performance in real-world machine learning applications.

Installation

Pip

# clone project
git clone https://github.com/YourGithubName/your-repo-name
cd your-repo-name

# [OPTIONAL] create conda environment
conda create -n myenv python=3.9
conda activate myenv

# install pytorch according to instructions
# https://pytorch.org/get-started/

# install requirements
pip install -r requirements.txt

Conda

# clone project
git clone https://github.com/YourGithubName/your-repo-name
cd your-repo-name

# create conda environment and install dependencies
conda env create -f environment.yaml -n myenv

# activate conda environment
conda activate myenv

How to run

Train model with default configuration

# train on CPU
python src/train.py trainer=cpu

# train on GPU
python src/train.py trainer=gpu

Train model with chosen experiment configuration from configs/experiment/

python src/train.py experiment=experiment_name.yaml

You can override any parameter from command line like this

python src/train.py trainer.max_epochs=20 data.batch_size=64

To Visualise in Voxel51

• Open notebooks/voxel_analysis.ipynb notebook.
• Modify paths of image and masks paths.
• Run the cell with :

view = dataset.sort_by("iou", reverse=True)
session = fo.launch_app(view , port = 5051)

For Analysis in Neptune

• Open configs/logger/neptune.yaml file.
• Enter your api key.
• Run the models.
• Open Neptune, loggings will be there.
• To change the name of experiment , open configs/experiment/building_seg.yaml file and modefy the "experiment_name".