project details

Food Vision

A multiclass food image classifier with command line and desktop interfaces.

PythonTensorFlowKerasPandasNumPyTkinter

#overview

Food Vision explores multiclass image classification in the concrete domain of food recognition. A convolutional neural network learns visual features from labeled examples and predicts the category of an unseen food image. The repository treats the model as part of a usable application rather than stopping at a training notebook.

Both command line and graphical paths make inference accessible after training. Supporting scripts automate repeated setup and execution work, while charts and tabular tooling help inspect the model's behavior. The project is a practical introduction to the full loop of preparing data, training, evaluating, saving, and serving a computer vision model.

#install and run

Prepare the environment

python -m venv env
pip install -r requirements.txt

Run the CLI or desktop app

python prediction.py --model 1 --food pizza
python GUI.py

#architecture and features

TensorFlow and Keras define and train the convolutional network. NumPy and Pandas support data preparation and analysis, while Matplotlib visualizes training and evaluation results. Saved model artifacts separate the expensive training phase from normal inference.

The inference layer loads a trained model, applies the same image resizing and normalization expected during training, and maps output probabilities back to class names. Tkinter provides the desktop interface, while the CLI offers a simpler path suitable for scripts or environments without a graphical session.

#inside the project

Food Vision model accuracy chart — Training and validation accuracy

Food Vision model loss chart — Training and validation loss

Food Vision command line prediction result — Command line prediction

Food Vision desktop prediction interface — Desktop interface

#using the project

01Clone the repository, create a Python environment, and install the pinned machine learning dependencies. Follow the repository's data preparation instructions if retraining. Otherwise, obtain or generate the expected saved model artifact.
02Run the training workflow to fit the classifier and inspect the resulting metrics and plots. Keep preprocessing identical across training and validation so evaluation reflects the model rather than a mismatched input pipeline.
03For prediction, supply an image through the CLI or open it in the desktop interface. The application prepares the image, runs inference, and displays the predicted class. Treat the result as a model estimate: performance depends on the training classes and how closely new images resemble that data.

view source on GitHub