Caltech 101 Classification

Description

This project implements a deep learning model for image classification on the Caltech 101 dataset. The model is trained using PyTorch and monitored using Weights & Biases (wandb).

Project Structure

Caltech_classification_fine_tuning_AlexNet/
├── src/                          
│   ├── inference.py              
│   ├── preprocess.py
│   ├── train.py              
│   ├── utils.py                  
│   └── config.py                 
├── models/                       
│   └── model_v1.pkl              
├── data/                         
│   └── schemas/                  
├── tests/                        
│   └── test_inference.py         
├── api/                          
│   └── app.py                    
├── deployment/                   
│   ├── Dockerfile                                             
│   └── docker-compose.yml                     
├── monitoring/                   
│   ├── metrics.py                
│   └── alerts.py                 
├── .github/workflows/            
│   └── deploy.yml                
├── requirements.txt             
├── README.md                    , 
└── .env.example                  

Installation

To run this project, you need to install the Weights & Biases library (wandb). You can install it via pip:

pip install wandb

You also need to clone the dataset repository inside the data folder. Run the following command in your terminal:

git clone https://github.com/MachineLearning2020/Homework2-Caltech101.git

Usage

Once you have installed the required libraries and cloned the dataset repository, change the directory to the src folder, you can use the following command-line arguments to customize the training process:

python caltech_101_kaggle_v4.py --epoch NUM_EPOCHS --batch_size BATCH_SIZE --lr LR --momentum MOMENTUM --weight_decay WEIGHT_DECAY --fine_tune_mode PRETRAINED --fine_tune_setting FINE_TUNE_SETTING

• --epoch: Number of epochs for training (default is 30).
• --batch_size: Batch size for training (default is 16).
• --lr: Learning rate (default is 0.001).
• --momentum: Momentum value (default is 0.9).
• --weight_decay: Weight decay (default is 1e-5).
• --fine_tune_mode: Boolean value indicating whether to use fine-tuning or not (default is True).
• --fine_tune_setting: Fine-tuning setting (default is "all").you may choose this parameter among all, classifier and features

Note: You can omit any arguments to use their default values.

Example

Here's an example command to start training:

python caltech_101_kaggle_v4.py --epoch 10 --batch_size 32 --lr 0.001 --momentum 0.9 --weight_decay 0.0001 --fine_tune_mode True --fine_tune_setting "setting_1"

This command will train the model for 10 epochs with a batch size of 32, learning rate of 0.001, momentum of 0.9, weight decay of 0.0001, using fine-tuning mode with setting "setting_1".

Results

Acknowledgments

• This project uses the Caltech 101 dataset.
• Weights & Biases (wandb) is used for experiment tracking and visualization.