Project Phases

Imports

• The project begins by importing essential libraries such as OpenCV, NumPy, and PIL to support image handling, resizing, and array manipulation.

• TensorFlow’s MobileNetV2 model, along with its preprocessing and decoding utilities, is imported to enable deep-learning inference using pretrained ImageNet weights.

• Streamlit is included to provide a clean and interactive web interface for running the image classifier in a browser.

Model Loading

• A dedicated function loads the MobileNetV2 model with pretrained ImageNet weights, ensuring the application has immediate access to a high-performance, lightweight neural network.

• The function returns the initialized model so that it can be reused throughout the application without repeated loading.

Preprocessing Step

• The preprocessing function converts uploaded images from PIL format into NumPy arrays for compatibility with OpenCV and TensorFlow.

• Images are resized to 224×224 pixels, matching the input dimensions expected by MobileNetV2.

• Pixel values are normalized using the model’s required preprocessing method, ensuring that inference matches the network’s training conditions.

• The image array is expanded along a new batch dimension so it can be passed into the model as a proper input tensor.

Model Prediction Process

• A classification function handles the full prediction cycle by preprocessing the image and passing it through the MobileNetV2 model.

• The model outputs raw probabilities that are then decoded into human-readable class labels and confidence scores.

• The function gracefully handles errors, ensuring that any issues during classification are surfaced clearly in the Streamlit interface.

Streamlit App Interface

• The main function configures the Streamlit page with a title, icon, and centered layout to create a simple and appealing UI.

• A short description is displayed to explain the purpose of the application and guide users through the workflow.

Cached Model Loader

• Streamlit’s resource caching mechanism is used to load the model once and reuse it across multiple interactions, significantly improving performance and reducing startup time.

File Upload Section

• A file uploader widget allows users to submit images in JPG or PNG formats directly through the browser.

• Once an image is uploaded, it is immediately displayed on the page to confirm successful submission.

Image Display and Interaction

• The interface provides a “Classify Image” button that triggers the prediction process only when pressed, giving users control over when inference begins.

• A loading spinner is shown during analysis to indicate that the model is actively processing the image.

Prediction Output

• After classification, the application displays a “Predictions” header followed by the top three predicted object labels and their corresponding confidence percentages.

• Each prediction is presented in a clean, readable format to help users interpret what the model recognized in their image.

Application Execution

• A standard Python entry point ensures the Streamlit app runs only when the file is executed directly, keeping the code clean and modular.