Skip to content

8. Web Application

The project includes a lightweight, built-in FastAPI web application with a vanilla JavaScript frontend. It is designed to be a complete dashboard for your classification model, allowing you to run predictions, trigger training jobs, evaluate the model, and interact with the optional memory layer.

Architecture Overview

The webapp lives entirely in the webapp/ directory: - server.py: The FastAPI backend handling REST API endpoints. - static/index.html: A single-page application (SPA) providing the user interface. - static/app.js: Vanilla JS logic (no framework like React or Vue required). - static/style.css: Styling and responsive design.

The architecture is built for simplicity. When a long-running process (like training or evaluation) is triggered, the backend uses a JobRunner component that spawns a subprocess (e.g., python -m src.train). It captures the stdout and stderr in real-time, allowing the frontend to poll for progress updates (roughly every 1.2 seconds) and display dynamic progress bars.

Starting the Web Server

To start the server, activate your virtual environment and run the main entry point:

# 1. Activate the environment
source .venv/bin/activate

# 2. Run the server
python webapp/server.py

By default, the server listens on http://localhost:8000. Open this URL in your web browser to access the dashboard.

Dashboard Features

The webapp is organized into tabs, each serving a distinct workflow:

1. Predict

Upload any image (Cattle or Buffalo) to test the active model. - The UI fetches the latest model checkpoint from outputs/checkpoints/. - Displays the predicted species (Cattle or Buffalo). - Displays the predicted breed. - Shows confidence scores.

Note on Model Cache: The ModelBox class in the backend caches the model weights in memory. If a new training job completes and overwrites the checkpoint, ModelBox detects the modified timestamp (mtime) and automatically invalidates the cache, loading the fresh model on the next prediction.

2. Train

Launch a training job directly from your browser. - Uses JobRunner to spawn python -m src.train. - Real-time progress bars and loss metrics are extracted from the subprocess standard output and displayed dynamically. - Automatically stops polling when the job signals completion.

3. Evaluate

Run the evaluation suite on the current checkpoint. - Spawns python -m src.evaluate. - Generates detailed confusion matrices and per-class metrics.

4. Memory

Interact with the Mem0 vector store. - A chat-like interface to store facts, retrieve context, or query the LLM memory. - Useful if you are extending the classifier with an AI assistant.

5. Debug

Technical details for developers. - View raw subprocess logs. - System information (CUDA availability, RAM, CPU). - Inspect currently stored checkpoints and exports in the outputs/ directory.

API Endpoints

The webapp can also be used headlessly via its REST API. Some of the core endpoints include:

  • POST /api/predict: Upload an image file for classification.
  • POST /api/train/start: Start a training job.
  • GET /api/train/status: Get current training progress and logs.
  • POST /api/evaluate/start: Start an evaluation job.
  • GET /api/evaluate/status: Get evaluation progress.

For a full list of payloads and schemas, refer to the API Reference.