# YOLO Studio YOLO Studio is a desktop developer tool for training, organizing, and remotely deploying YOLO computer vision models. ## Project Overview This project provides: - Local dataset management and dataset-building workflows - Configurable YOLO training with live metrics and logs - Discovery integrations for Roboflow Universe, Hugging Face Hub, Kaggle, and Open Images - Remote deployment and testing against Jetson/Raspberry Pi edge devices - Real-time camera inference with session capture - Offline evaluation with analytics dashboards - Cross-run analytics and experiment comparison - Project system for organizing assets under named roots ## Setup 1. Create and activate a Python 3.10+ virtual environment. 2. Install dependencies: \`\`\`bash pip install -r requirements.txt \`\`\` 3. Launch the app: \`\`\`bash python main.py \`\`\` ## Architecture \`\`\`text +-------------------------- YOLO Studio (PyQt6) ---------------------------+ | | | +---------------- UI Layer ----------------+ +------ Widgets -------+ | | | train | dataset | discover | remote | | log | chart | cards | | | | camera | evaluate | analytics | +----------------------+ | | +------------------------------------------+ | | | | | v | | +------------------ Core Layers ----------------------------+ | | | models | services | workers | projects | | | +----------------------------------------------------------+ | | | | | v | | +------------------------- Integrations ---------------------------+ | | | ultralytics | roboflow | huggingface_hub | websocket devices | | | +------------------------------------------------------------------+ | +--------------------------------------------------------------------------+ Edge Device Agent (Jetson/RPi): edge/jetson_agent.py \`\`\` ## Code Layout - \`core/models/\`: SQLAlchemy entities and DB session/bootstrap logic - \`core/services/\`: dataset and remote-device service classes - \`core/workers/\`: QThread background worker classes - \`ui/windows/\`: top-level window classes - \`ui/styles/\`: theme and styling modules - \`ui/tabs/\`, \`ui/widgets/\`: feature tabs and reusable UI components - Compatibility shims remain at legacy module paths (\`core/database.py\`, etc.) for safer import migration. ## Usage Guide ### Training Tab - Configure model architecture, dataset, and hyperparameters - Start/stop training jobs with live metrics and logs - Save successful runs and manage trained weights ### Dataset Tab - Browse datasets stored in SQLite - Build datasets from local images and generate YOLO metadata - Review and export saved models Note: Project scope filters the dataset library and saved models. ### Discover Tab - Search and download datasets from Roboflow Universe - Search and download models/datasets from Hugging Face Hub - Search Kaggle datasets/competitions and download datasets - Download Open Images subsets with class filters Note: Imported assets can be attached to the active project. ### Remote Devices Tab - Register edge devices - Deploy model weights remotely - Run remote inference tests and store result metrics ### Camera Tab - Run real-time webcam inference with overlays - Record frames, save snapshots, export detections CSV - Session metadata stored in the database ### Evaluate Tab - Offline evaluation for datasets or image folders - Confusion matrix, PR curves, and confidence curves - Export reports as PDF + JSON ### Analytics Tab - Cross-run history, loss overlays, and per-class heatmaps - Compare runs and export dashboard ZIP ### Project System - Create/open/duplicate/delete projects - Store project config in \`project.yaml\` - Project selector filters all tabs ## Edge Agent Setup 1. Copy \`edge/jetson_agent.py\` and \`edge/agent_config.yaml\` to the device. 2. Install Python dependencies on the device. 3. Configure \`agent_config.yaml\`. 4. Start the agent and verify connectivity from YOLO Studio. ## Troubleshooting - Ensure Python version is 3.10+. - Confirm \`config.json\` contains valid API tokens when using external services. - Verify CUDA/device drivers for GPU-enabled edge inference. - Check application and edge agent logs for runtime errors. - On Linux desktops without working Wayland or XCB dependencies (e.g. missing \`libxcb-cursor0\`), YOLO Studio falls back to Qt's \`minimal\` platform. You can override with \`QT_QPA_PLATFORM\` if you want to force a specific backend. ## Current Status The application scaffold and core feature modules are implemented, including: - SQLAlchemy-backed metadata models and persistence helpers - Training, dataset, discover, and remote-device GUI tabs - QThread workers for long-running operations - WebSocket remote manager and edge agent protocol implementation