CIFAR10-optimal-and-accurate-solutions

Overview

This repository contains two Jupyter notebooks that provide solutions for the CIFAR-10 dataset using two different models: ResNet-9 and Vision Transformer (ViT). The focus is on offering users a choice between a lightweight model with reasonable accuracy and a more complex model that achieves higher accuracy.

Models

1. ResNet-9

   • Description: ResNet-9 is a lightweight deep residual network designed for fast image classification. It consists of 9 layers and is optimized for performance on less complex tasks, making it suitable for scenarios where computational resources are limited.
   • Accuracy: Achieves approximately 92.5% accuracy on the CIFAR-10 dataset.
   • Model Weight: Lightweight, making it ideal for deployment in resource-constrained environments.

2. Vision Transformer (ViT)

   • Description: Vision Transformer leverages self-attention mechanisms to process images. It has shown superior performance on various image classification tasks, especially when trained on large datasets.
   • Accuracy: Achieves an impressive 97.9% accuracy on the CIFAR-10 dataset.
   • Model Weight: Heavier than ResNet-9, requiring more computational resources but providing significantly better accuracy.

Comparison

Model	Accuracy	Model Weight
ResNet-9	0.925	Lightweight
ViT	0.979	Heavier

Key Differences

• Performance: While ResNet-9 provides a good balance of speed and accuracy, ViT excels in terms of accuracy at the cost of increased model complexity and resource requirements.

• Use Cases:

  • Use ResNet-9 when you need a fast solution that can run efficiently on devices with limited computational power.
  • Choose ViT when the highest possible accuracy is required and you have access to sufficient computational resources.