Project 4: Application of Computer Vision
Overview:
This project explores the application of Convolutional Neural Networks (CNNs) for image classification using the CIFAR-10 dataset with the Keras library. The primary goal is to demonstrate the effectiveness of CNNs in recognizing and classifying images.
Key Components:
CNN Architecture: Utilizes convolutional layers, ReLU activations, pooling layers, and fully connected layers to extract and classify image features. Dataset: The CIFAR-10 dataset, consisting of 60,000 32x32 color images across 10 classes. Frameworks: Keras for building and training models and TensorFlow as the backend for scalability and deployment.
Methodology:
Data Acquisition: Loading and normalizing the CIFAR-10 dataset. Model Implementation: Constructing a CNN using Keras' Sequential API, including convolutional, pooling, and dense layers. Training: Employing gradient-based optimization (Adam optimizer) and learning rate scheduling to train the model. Performance Evaluation: Assessing model performance using metrics like accuracy, loss, and F1 score; visualizing training and validation accuracy/loss.
Results:
The model demonstrates significant improvement in training accuracy over epochs. Validation metrics indicate the need for further tuning to prevent overfitting or underfitting.
Conclusion:
The project successfully showcases the power of CNNs in image classification tasks and provides insights into model optimization techniques for better performance on the CIFAR-10 dataset.
Demo: https://youtu.be/UK1qTgCY0Ug?si=jiA_A-UOCs6aSa2p
| Pipeline Step | What I Did | Azure Equivalent |
|---|---|---|
| Data Ingestion | Loaded CIFAR-10 dataset from Keras | Azure Data Factory (linked dataset) |
| Processing | Normalized, augmented, and reshaped image data | Azure Databricks (Python pipeline) |
| Modeling/Analysis | Built and trained CNN using Keras/TensorFlow | Azure Databricks ML Runtime |
| Evaluation | Measured accuracy, loss, F1 score | DBT (metrics layer) |
| Automation | Could simulate with scheduled training script | Azure Logic Apps / Pipelines |
| Versioning | Used GitHub for model updates and performance tracking | Azure DevOps Git |