Brand Logo Classification

A PyTorch-based computer vision project for brand logo classification across 7 classes (Apple, Google, Samsung, Facebook, WhatsApp, Messenger, Mozilla). Implements multiple state-of-the-art deep learning architectures with comprehensive training utilities, data augmentation, and hyperparameter tuning capabilities.

Features

• Multiple Model Architectures: CNN, MLP, ResNet18, SENet, Vision Transformer (ViT), and Ensemble methods
• Advanced Training Pipeline: Configurable training with data augmentation, learning rate scheduling, and early stopping
• Class Imbalance Handling: Weighted loss functions for imbalanced datasets
• Hyperparameter Tuning: Automated hyperparameter optimization using Optuna
• Data Augmentation: Comprehensive augmentation strategies for improved generalization
• Kaggle Integration: Inference utilities for competition submissions
• CI/CD Integration: Automated testing and linting with GitHub Actions
• Code Quality: Enforced code formatting with Black, import sorting with isort, and type checking with mypy

Project Structure

computer-vision/
├── models/                    # Neural network architectures
│   ├── CNN.py                # Convolutional Neural Network
│   ├── MLP.py                # Multi-layer Perceptron
│   ├── ResNet.py             # ResNet18 implementation
│   ├── SENET.py              # Squeeze-and-Excitation Network
│   ├── VIT.py                # Vision Transformer
│   ├── Ensemble.py           # Ensemble methods for model combinations
│   └── checkpoints/          # Saved model checkpoints
├── utils/                     # Training and utility functions
│   ├── train_model.py        # Main training script
│   ├── tune_hyperparameters.py  # Hyperparameter optimization
│   ├── augment_data.py       # Data augmentation utilities
│   ├── dataloader.py         # Dataset loading and preprocessing
│   └── kaggle_inference.py   # Kaggle submission generation
├── config/                    # Model configuration files (YAML)
│   ├── cnn.yaml
│   ├── mlp.yaml
│   ├── resnet.yaml
│   ├── senet.yaml
│   └── vit.yaml
├── data/                      # Dataset directory
│   ├── train/                # Training images (organized by class)
│   ├── test/                 # Test images for Kaggle submission
│   └── train_labels.csv      # Training labels
├── weights/                   # Pretrained model weights
├── tests/                     # Unit and integration tests
├── notebooks/                 # Jupyter notebooks for experiments
├── .github/workflows/         # CI/CD configuration
│   ├── test.yml              # Automated testing
│   └── lint.yml              # Code quality checks
├── Makefile                   # Development commands
└── TRAINING_GUIDE_95PERCENT.md  # Guide to achieving 95%+ accuracy

Running train_colab.ipynb on Google Colab

1. Open Google Colab
   Go to Google Colab in your browser.

2. Load the Notebook

   • Click File > Upload Notebook if you have train_colab.ipynb locally, and select your file.
   • Alternatively, if the notebook is in a GitHub repo, choose File > Open notebook > GitHub and paste the repository link.
   • You can also drag and drop the notebook into the Colab interface.

3. Mount Google Drive (optional but recommended)
   If your data or weights are stored in Google Drive, insert and run the following cell:

   from google.colab import drive
   drive.mount('/content/drive')

   This allows you to access files in your Drive.

4. Check/Set Up Dependencies

   • Make sure all necessary packages are installed in the first notebook cells (e.g., !pip install -r requirements.txt).
   • You can upload your requirements.txt or install packages one by one using pip inside notebook cells, such as:

     !pip install torch torchvision optuna

5. Upload or Access Data

   • If data is in Drive, set data paths appropriately (e.g., /content/drive/MyDrive/yourfolder/data/).
   • If you need to upload from your computer, use:

     from google.colab import files
     uploaded = files.upload()

6. Start Training

   • Read through and run the notebook cells in sequence by clicking the "Play" button on the left of each code cell, or select Runtime > Run all.
   • Monitor outputs for warnings/errors.
   • Training progress, loss, and accuracy will be shown in the notebook outputs.

7. Saving Outputs

   • After training, save your models or submission files to your Google Drive or download to your machine.

Tips:

• Restart your runtime if you change pip dependencies: Runtime > Restart runtime.
• Use a GPU for faster training: Runtime > Change runtime type > Hardware accelerator > GPU.

This process will let you easily load and run the train_colab notebook on Colab and start training your models end-to-end.