Weather Image Classification

This repository provides an end-to-end pipeline for image classification using PyTorch. It includes functionalities for:

• Loading and preprocessing images.
• Splitting datasets into training, validation, and test sets.
• Defining custom PyTorch Dataset classes.
• Training a convolutional neural network (CNN) model.

Data Set

The data set of the 11 different weather conditions is taken from Kaggle.

Motivations

Personally, I think that climate change is one of the most pressing social issue which is impacting so many people around the world thus, I hope to make use of AI to reduce the impact of climate change in people's lives. Hence this will be one of the many projects that I'll do to not only to solidify my knowledge in AI but also to play a part in societal issues like climate change.

Features

• Data Handling:
  Uses PIL and torchvision.transforms for image loading and preprocessing.
  Splits data into train/validation/test sets using sklearn.model_selection.train_test_split.

• Model Training and Evaluation:
  Implements training loops with torch.optim and nn.CrossEntropyLoss.
  Utilizes progress bars via tqdm to monitor training progress.
  Supports GPU acceleration if available.

Process:

1. Visualizing given data using matplotlib
2. Splitting the data using sklearn train_test_split function
3. Train the data using a CNN with the usage of the Pytorch framework.

Future plans:

With the current model running at an average of 50% accuracy rate, I hope to increase it to 80-90% accuracy consistently by enchancing and fine-tuning the model.

Prerequisites

• Python 3.7+
• PyTorch and TorchVision
• PIL (via Pillow)
• scikit-learn
• tqdm
• matplotlib
• pandas
• numpy

You can install these dependencies with:

pip install torch torchvision pillow scikit-learn tqdm matplotlib pandas numpy