Image Classification using DL Methods Version 2

[UTSAVS26]

Deep Learning Simplified Repository - New Issue Template

🔴 Project Title : Image Classification using Convolutional Neural Networks (CNN)
🔴 Aim : To implement and compare various convolutional neural network (CNN) architectures for image classification tasks using the CIFAR-10 and MNIST datasets. The goal is to evaluate model performance and accuracy based on several algorithms.
🔴 Dataset :

• CIFAR-10 Dataset: Link to CIFAR-10 Dataset
• MNIST Dataset: Link to MNIST Dataset

🔴 Approach :
The goal of this project is to use 3-4 different CNN algorithms to implement the models, train them on the CIFAR-10 and MNIST datasets, and then compare all the algorithms' performance by evaluating their accuracy scores.

• Conduct Exploratory Data Analysis (EDA) before model creation, including data visualization, normalization, and preprocessing.
• Implement various CNN architectures such as LeNet-5, MobileNet, ResNet50, Simple CNN, and VGG16, and evaluate their performance on both CIFAR-10 and MNIST datasets.
• The final goal is to determine which algorithm performs the best for each dataset based on accuracy scores.

---

✅ To be Mentioned while taking the issue:

• Full Name: Utsav Singhal
• GitHub Profile Link: UTSAVS26
• Email ID: utsavsinghal26@gmail.com
• Participant ID (if applicable):
• What is your participant role? SWOC

Approach for this Project:

1. Dataset Preparation:

   • Load CIFAR-10 and MNIST datasets.
   • Normalize pixel values and apply resizing/augmentation where needed.
   • Split data into training, validation, and test sets.

2. Exploratory Data Analysis (EDA):

   • Visualize images and check class distribution to ensure balanced datasets.
   • Identify any data issues that may affect model performance.

3. Model Implementation:

   • Implement 3-4 CNN architectures:
     • LeNet5_Model: Simple model for MNIST.
     • MobileNet_Model: Efficient architecture for both datasets.
     • ResNet50_Model: Deeper model for CIFAR-10.
     • VGG16_Model: Complex model for CIFAR-10.

4. Model Training:

   • Train each model using appropriate optimizers and loss functions.
   • Tune hyperparameters and implement early stopping to prevent overfitting.

5. Model Evaluation:

   • Evaluate models using accuracy, precision, recall, and F1-score.
   • Visualize results with confusion matrices and performance curves.

6. Comparison & Conclusion:

   • Compare models based on accuracy scores and performance.
   • Recommend the best model for each dataset.

7. Documentation:

   • Provide a detailed README.md with model summaries, visualizations, and conclusions.
   • List dependencies in requirements.txt.

This approach will allow for an efficient comparison of CNN models to determine the best fit for MNIST and CIFAR-10 image classification tasks.

---

Happy Contributing 🚀

All the best. Enjoy your open source journey ahead. 😎

[github-actions]

Thank you for creating this issue! We'll look into it as soon as possible. Your contributions are highly appreciated! 😊

[abhisheks008]

Similar problem statement is already present here in this repository. Why should I go ahead with your project and remove the existing one, can you elaborate that?

Here is link for your reference, https://github.com/abhisheks008/DL-Simplified/tree/main/MNIST%20Digit%20Classification%20using%20Neural%20Networks

@UTSAVS26

[UTSAVS26]

My project offers significant value and expansion to the existing problem statement in the repository for the following reasons:

1. Multiple CNN Architectures: Unlike the current project, which focuses on a basic Artificial Neural Network (ANN) and Convolutional Neural Network (CNN) comparison, my project includes the implementation and evaluation of multiple CNN architectures such as LeNet-5, MobileNet, ResNet50, and VGG16. By comparing these advanced architectures, the project provides a much deeper understanding of their performance and suitability for image classification tasks.

2. Evaluation on Two Datasets: My project goes a step further by evaluating models on both the CIFAR-10 and MNIST datasets, which allows for a broader assessment of model performance across datasets with different characteristics (grayscale vs. color images, smaller vs. larger image sizes). This comparison adds more depth and applicability to the findings.

3. Enhanced Model Evaluation: In addition to accuracy, my project evaluates models based on precision, recall, and F1-score, and includes confusion matrices for a more detailed and insightful analysis of each model’s performance. This multi-metric approach provides a clearer picture of how well the models perform in various scenarios.

4. Data Preprocessing and Hyperparameter Tuning: The project emphasizes Exploratory Data Analysis (EDA), hyperparameter tuning, and early stopping, ensuring that the models are well-optimized and free from overfitting. These steps contribute to more reliable and high-performing models.

In conclusion, my project does not replace the existing one, but rather complements it by introducing multiple CNN models, expanding the dataset scope, and offering a more thorough and detailed evaluation. It provides a deeper and more comprehensive approach to image classification using CNNs, making it a valuable addition to the repository.

@abhisheks008

[abhisheks008]

I have updated the issue title. Please follow the same issue title for the project folder as well as for the pull request title.

Issue assigned to you @UTSAVS26

[github-actions]

Hello @UTSAVS26! Your issue #995 has been closed. Thank you for your contribution!