Issue: #102 – Increase the Count of Less Represented Classes

[Krishna200608]

Issue: #102

Summary

This Pull Request addresses class imbalance in the astronomy image dataset by
demonstrating data augmentation techniques for under-represented classes.

The goal is to improve class representation during training without increasing
dataset size on disk.

Work Done

• Analyzed class distribution in the training dataset
• Identified under-represented classes automatically
• Implemented lightweight image augmentation techniques:
  • Horizontal flipping
  • Small-angle rotations
  • Brightness and contrast adjustments
• Applied augmentation only to the training set
• Demonstrated augmentation in memory without writing new images to disk

Augmentation Strategy

• Augmentation is performed virtually (in-memory) to avoid dataset expansion
• A bounded number of augmented samples is generated per minority class
• This mirrors real-world training pipelines where augmentation is applied
  dynamically during model training

Key Outcome

• Minority classes receive additional virtual samples during training
• Class imbalance is reduced without affecting validation or test data
• No additional storage or permanent dataset modification is required

Implementation Notes

• The implementation is added as a new section in the notebook:
  participants/IIT2023139/notebook_v3.ipynb
• No dataset files or augmented images are uploaded to the repository
• The solution is safe for Kaggle environments and fully reproducible

Environment

• Platform: Kaggle Notebook
• OS: Windows 11
• Editor: VS Code

This PR addresses only Issue #102 and follows all OpenCode contribution
guidelines.

[OpenGitBot]

Hey @Krishna200608

Thanks for opening this PR 🚀. Mentor will review your pull request soon and till then, keep contributing and stay calm.

Thanks for contributing in OpenCode'25 ✨✨!