INSTALLATION.md

Installation Guide for FoodViT

Prerequisites

• Python 3.8 or higher
• pip package manager
• At least 4GB RAM (8GB recommended)
• GPU support optional but recommended for faster inference

Installation Steps

1. Clone or Download the Project

Make sure you have all the project files in your directory:

• app.py - Main application
• predict.py - Command line tool
• config.py - Configuration
• requirements.txt - Dependencies
• model/bestViT_PT.pth - Trained model
• All utility and interface files

2. Create a Virtual Environment (Recommended)

# Create virtual environment
python -m venv foodvit_env

# Activate virtual environment
# On Windows:
foodvit_env\Scripts\activate
# On macOS/Linux:
source foodvit_env/bin/activate

3. Install Dependencies

# Install PyTorch first (choose appropriate version for your system)
# For CPU only:
pip install torch torchvision --index-url https://download.pytorch.org/whl/cpu

# For CUDA (if you have NVIDIA GPU):
# pip install torch torchvision --index-url https://download.pytorch.org/whl/cu118

# Install other dependencies
pip install -r requirements.txt

4. Troubleshooting Dependency Issues

If you encounter dependency conflicts, try this step-by-step approach:

# 1. Install core dependencies first
pip install torch torchvision
pip install transformers==4.28.0
pip install huggingface-hub==0.15.1
pip install accelerate==0.20.3

# 2. Install image processing libraries
pip install Pillow opencv-python albumentations

# 3. Install Gradio
pip install gradio==3.35.2

# 4. Install other utilities
pip install numpy scikit-learn datasets

5. Alternative: Use Conda

If you prefer conda:

# Create conda environment
conda create -n foodvit python=3.9
conda activate foodvit

# Install PyTorch
conda install pytorch torchvision -c pytorch

# Install other packages
pip install transformers==4.28.0 huggingface-hub==0.15.1
pip install gradio==3.35.2
pip install -r requirements.txt

Testing the Installation

1. Run Basic Tests

python simple_test.py

This should show all tests passing.

2. Test the Web Interface

python app.py

Then open your browser to http://localhost:7860

3. Test Command Line Tool

# Test help
python predict.py --help

# Test with a sample image (if you have one)
python predict.py path/to/your/image.jpg

Common Issues and Solutions

Issue: "cannot import name 'split_torch_state_dict_into_shards'"

Solution: This is a version compatibility issue. Try:

pip uninstall huggingface-hub transformers accelerate
pip install huggingface-hub==0.15.1 transformers==4.28.0 accelerate==0.20.3

Issue: CUDA/GPU not working

Solution:

1. Check if you have NVIDIA GPU
2. Install appropriate CUDA version
3. Install PyTorch with CUDA support
4. Or set device to 'cpu' in config.py

Issue: Model file not found

Solution: Ensure model/bestViT_PT.pth exists in the project directory.

Issue: Memory errors

Solution:

1. Close other applications
2. Use CPU instead of GPU
3. Reduce batch size in configuration

System Requirements

Minimum Requirements

• Python 3.8+
• 4GB RAM
• 500MB disk space

Recommended Requirements

• Python 3.9+
• 8GB RAM
• NVIDIA GPU with CUDA support
• 1GB disk space

Verification

After successful installation, you should be able to:

1. ✅ Run python simple_test.py without errors
2. ✅ Start the web interface with python app.py
3. ✅ Use command line tool with python predict.py --help
4. ✅ Upload images and get predictions in the web interface

Getting Help

If you encounter issues:

1. Check the error messages carefully
2. Ensure all dependencies are installed correctly
3. Try the troubleshooting steps above
4. Check if your Python version is compatible
5. Verify the model file exists and is not corrupted

Next Steps

Once installation is complete:

1. Web Interface: Run python app.py and visit http://localhost:7860
2. Command Line: Use python predict.py for batch processing
3. Customization: Edit config.py to modify settings
4. Development: Use the modular structure for extending functionality