README.md

---
license: mit
language:
- en
- pt
metrics:
- accuracy
- bertscore
pipeline_tag: image-classification
tags:
- code
---

# Galaxy10 SDSS Dataset CNN TF Model

This model contains a Convolutional Neural Network (CNN) ensemble model trained on the Galaxy10 SDSS Dataset available in the astroNN library.

## Dataset
The Galaxy10 SDSS Dataset is a collection of astronomical images categorized into 10 classes of galaxies. The dataset has undergone data augmentation to ensure class balance.

To access the dataset:
- Visit the [astroNN Galaxy10 SDSS Documentation](https://astronn.readthedocs.io/en/stable/galaxy10sdss.html)

## Models Used
The ensemble model comprises the following pre-trained CNN architectures:
- AlexNet
- DenseNet121
- ResNet50
- EfficientNet-V2-M

## Ensemble Learning Technique
The ensemble model employs Mean Voting to aggregate predictions from individual models. This technique combines the output probabilities or predictions from each model and calculates the mean for final classification.

## Results
After training and evaluation on the Galaxy10 SDSS Dataset, using the test dataset, the ensemble model achieved the following metrics:
- Accuracy: 0.9175
- Precision: 0.9203
- Recall: 0.9168
- F1-score: 0.9175
- Loss: 0.4963


## Training
The models were trained using PyTorch, leveraging transfer learning for some architectures and training from scratch for others. The ensemble learning technique was applied post-training using the mean voting strategy.

## Requirements
- PyTorch v2.1.0
- CUDA v11.8 (cu118)

## Acknowledgments
- Credits to the astroNN team for providing the Galaxy10 SDSS Dataset and related documentation.
- The pre-trained model architectures used are courtesy of Pytorch.

Feel free to reach out for any queries, issues, or improvements related to this model repository.