Data

Tools

Indexes

Applications

Experiments

Awesome

An open API service indexing awesome lists of open source software.

https://github.com/star-rein/onnx-detect

A general object detection application supporting ONNX models, image/video/camera inputs, with one-shot and real-time inference modes, and an intuitive GUI.
https://github.com/star-rein/onnx-detect

gui object-detection onnx onnx-runtime pyqt6 pyqt6-desktop-application yolo

Last synced: about 1 month ago
JSON representation

A general object detection application supporting ONNX models, image/video/camera inputs, with one-shot and real-time inference modes, and an intuitive GUI.

Awesome Lists containing this project

README 

          
# ONNX Detect



![ONNX-Detect-Screenshot](./assets/screenshot_v1.3.3.png)



This is a modern desktop application built on **PyQt6**, **PyQt-Fluent-Widgets**, and **ONNXRuntime**, specifically designed for ONNX model object detection. It features a qframelesswindow-based Acrylic interface, supporting multiple input sources, flexible model management (built-in and custom), and robust camera control capabilities.



---



## 🌎 Language



* **[Read the original Chinese (δΈ­ζ–‡) README here](./README_zh.md)**



---



## πŸš€ Core Features



* **Modern UI Interface**:

    * Built with `PyQt6` and the `pyqt6-fluent-widgets` library, providing a smooth Fluent Design (WinUI) style.

    * Supports the Windows 11 Acrylic translucent background effect and allows users to customize the background color tone.

* **High-Performance Inference Backend**:

    * Utilizes `onnxruntime-gpu` as the core inference engine, prioritizing **NVIDIA GPU (CUDA)** for acceleration, with seamless fallback to CPU.

    * The UI automatically displays the currently used inference device (GPU or CPU).

* **Flexible Inference Modes**:

    * **One-Time Inference**: Supports image/video files and single-frame camera capture.

    * **Real-Time Inference**: Supports video files and live camera input with pause/resume functionality.

    * **Batch Inference (New)**:

        * Supports **queuing multiple files** (images/videos) for automated sequential processing.

        * Visual progress tracking for the entire batch task.

* **Powerful Model Management**:

    * **Built-in Models**: Automatically loads all YOLOv10 (n, s, m, l, x) ONNX models.

    * **Custom Models**: Loads arbitrary ONNX models via YAML config, with full support for custom class names and colors.

* **Batch Processing & Data Export**:

    * **Smart Naming**: Customizable filename templates (e.g., `{original}_{timestamp}_{index}`) for organized output.

    * **CSV Data Export**: Automatically generates CSV reports containing detailed detection data (classes, confidence, coordinates) for analysis.

    * **Auto-Indexing**: Smart output directory management with automatic folder creation and index incrementing.

* **Advanced Camera Control**:

    * Automatically detects system cameras with "Enable/Disable" switching.

    * **Resolution Settings**: Detects supported resolutions and allows user pre-selection.

* **Comprehensive File Handling**:

    * **Full support for Chinese (Unicode) paths**.

    * Supports multiple image save formats (JPG, PNG, BMP, TIFF).

* **Result Playback and State Management**:

    * Built-in side-by-side video player for inference results.

    * Robust application state machine to prevent user operation errors.

* **Other Features**:

    * "About" page, "Clear" function, and a mysterious Easter Egg.



---



## πŸ“œ Changelog



Click to Expand/Collapse



Version Update Log


V1.3.4 - November 25, 2025


    

        
  • Fixed font color issues in Windows 11 Dark Mode.
    • 

        
  • Optimized interface font consistency.
    • 

        
  • Simplified built-in models.
    • 



V1.3.3 - November 22, 2025


     

        
  • Architecture Refactoring: Main interface code fully decoupled.
    • 

        
  • Bug Fix: Fixed the bug where USB camera hot-plugging was not recognized.
    • 

        
  • Uisual Optimization: Optimized the display size of bounding boxes and labels under high resolution.
    • 



V1.3.2 - November 20, 2025


    

        
  • Camera Management Refactoring: Supports displaying real hardware names, parallel detection, and optimized initialization speed.
    • 

        
  • Device Refresh: Identify newly inserted cameras via the "Refresh Device List" menu without restarting.
    • 

        
  • Core Optimization: Adopted native Qt APIs for device information to ensure better compatibility.
    • 



V1.3.1 - November 20, 2025


    

        
  • 

    Core Refactoring: Added full support for YOLOv11/v8 model formats and Dynamic Shape inputs.
    • 

        
  • 

    Performance Boost: Implemented NumPy vectorized post-processing and Letterbox preprocessing for significantly higher FPS and accuracy.
    • 

        
  • 

    Stability Fix: Fixed application crashes caused by OpenCV NMS return type inconsistencies.
    • 

        
  • 

    UX Improvements: Optimized real-time FPS calculation (instantaneous) and implemented asynchronous model loading for smoother startup.
    • 



V1.3.0 - November 19, 2025


    

        
  • 

    New Batch Inference Mode: Supports multi-file queue processing and real-time progress display.
    • 

        
  • 

    New Export Settings: Supports custom filename templates (timestamp, index, model name, etc.).
    • 

        
  • 

    New CSV Data Export: Supports generating summary reports or individual data files for batch tasks.
    • 

        
  • 

    Optimized Batch Management: Supports smart output directory indexing and automatic folder creation.
    • 



V1.2.4 - November 16, 2025


    

        
  • Introduced confidence and IOU threshold sliders, allowing for dynamic adjustment.
    • 

        
  • Changed the logo and added an initialization page.
    • 



V1.2.3 - November 13, 2025


    

        
  • Switching the OpenCV camera API to MSMF on Windows to resolve stuttering issues with high-resolution cameras.
    • 

        
  • Optimizing User Experience: Initialization and Resource Reloading
    • 



V1.2.2 - October 31, 2025


    

        
  • Added "Camera Settings" feature, allowing selection of camera resolution.
    • 



V1.2.1 - October 30, 2025


    

        
  • Added a [Mysterious Easter Egg], triggered by a mysterious number.
    • 

        
  • Added "About" interface, displaying the update log.
    • 

        
  • Added "Clear" function, allowing the clearing of the output preview.
    • 

        
  • Full support for Chinese path file operations; provides multiple formats for image files when saving inference results.
    • 



V1.2.0 - October 29, 2025


    

        
  • Code structure refactoring, referencing the MVVM architecture and integrating high-level abstraction layers.
    • 

        
  • Fixed several potential user UI interaction bugs.
    • 



V1.1.1 - October 28, 2025


    

        
  • Added camera system: supports detection, selection, and enabling/disabling of cameras.
    • 

        
  • One-time and real-time inference modes support the camera as an input source.
    • 

        
  • Optimized UI state management to enhance user experience.
    • 



V1.1.0 - October 27, 2025


    

        
  • Added custom model loading, management, and configuration features.
    • 

        
  • Added custom theme color selection feature.
    • 



V1.0.0 - October 26, 2025


    

        
  • Initial release, resolved onnx-runtime compatibility issue preventing normal packaging of the executable with pyinstaller.
    • 

        
  • Used the PyQt6-fluent-widgets third-party library for UI beautification.
    • 

        
  • Supports the Acrylic interface effect under Win11.
    • 



V0.0.0 - October 26, 2025


    

        
  • Initial version, supports single YOLOv10 ONNX model loading and inference.
    • 

        
  • Provides one-time image/video inference and real-time video inference features.
    • 

        
  • Standard QT UI.
    • 



GitHub: Click to Visit



---



## πŸ› οΈ Installation and Running



### Option 1: (Recommended) Using the Packaged .exe File



1.  Download the latest `.exe` executable file from the [Releases](https://github.com/STAR-REIN/ONNX-Detect/releases) page of this repository and download the environment compressed package from the [cloud drive link](https://pan.baidu.com/s/1tn5E1JG5FpbbVukE9UkVGg?pwd=ntdn) (password: ntdn). If you **don't want to download environment compressed package from cloud disk**, [Releases-v1.3.4](https://github.com/STAR-REIN/ONNX-Detect/releases/tag/v1.3.4) includes three versions of envpackage, you can download directly via GitHub.

2.  Environment compressed package version descriptions:

    * No suffix: Complete version, compressed package size is about 1.18G; comes with CUDA and onnx-runtime-gpu environment. You only need a GPU that supports CUDA 12.x.x to use GPU inference. **Recommended for users with a GPU but no pre-installed CUDA environment.**

    * Lite_GPU: Simplified version, compressed package size is about 514MB; retains GPU support but requires the user to install the CUDA environment themselves. **Recommended for users with an existing CUDA environment.**

    * Lite_CPU: Simplified version, compressed package size is about 454MB; removes GPU support and only supports CPU inference. **Recommended for users without a GPU.**

3.  First, download and decompress the required environment package. Then, place the `.exe` file into the root directory of the decompressed folder and double-click to run.

4.  Ensure that your `models` folder and (optional) `custom_models` folder are in the same directory as the `.exe` file.

5.  Directly run the `.exe` file.



### Option 2: Running from Source Code



1.  **Clone the Repository**:

    ```bash

    git clone [https://github.com/STAR-REIN/ONNX-Detect.git](https://github.com/STAR-REIN/ONNX-Detect.git)

    cd ONNX-Detect

    ```



2.  **Create Conda Environment**:

    This project uses the `environment.yml` file to manage dependencies.

    ```bash

    conda env create -f environment.yml

    ```



3.  **Activate Environment**:

    ```bash

    conda activate pyqt6_package

    ```



4.  **Prepare Models**:

    * Download the built-in model compressed package from the [cloud drive link](https://pan.baidu.com/s/). (Cloud drive link to be updated)

    * Place your downloaded YOLOv10 `basic` and `enhance` ONNX model files into the `models` folder in the root directory.

    * (Optional) Configure the `custom_models` folder according to the instructions in the next section.



5.  **Run the Program**:

    ```bash

    python main.py

    ```



**Dependency Notes**:

* This project requires **Python 3.12.1**.

* GPU acceleration relies on `onnxruntime-gpu==1.19.0`. Please ensure your **NVIDIA driver** and **CUDA Toolkit** versions are compatible with ONNXRuntime. If your GPU is not supported, `onnxruntime-gpu` will automatically fall back to CPU mode.



---



## πŸ“– Usage Guide



1.  **Start the Program**: Run `.exe` or `python main.py`.

2.  **Select a Model**:

    * **Built-in Models**: Click "Built-in Model Selection" and choose a model from the dropdown menu. The program will load it automatically.

    * **Custom Models**: Click "External Model Management" -> "Load Custom Models" to load your YAML configuration. Then click "External Model Selection" to choose a model.

3.  **Select Inference Mode**:

    * **One-Time Mode**: For processing a single file or a single frame.

    * **Run-Time Mode (Real-Time)**: For processing video files or a live camera stream.

4.  **Select Input Source**:

    * **File**: Click the "Upload File" button and select an image or video. A preview will display after upload.

    * **Camera**:

        1.  (Optional) Click the "Camera Settings" button. While the **camera system is disabled**, pre-select a resolution for the camera you plan to use.

        2.  Click "Camera Selection" -> "Enable/Disable βœ•" to start the camera system.

        3.  Click "Camera Selection" again and select a detected camera from the list (e.g., "Camera 0").

        4.  The input preview area will now display the live camera feed.

5.  **Start Inference**:

    * Click the "Start Inference" (or "Start Real-Time Inference") button.

    * The button will change to "Stop Inference" during the process.

    * In real-time inference mode, the "Play/Pause" button can be used to pause/resume the inference thread.

6.  **View Results**:

    * **One-Time Mode**: Results will appear in the right-side "Inference Results" area. If it's a video, the player will load automatically after processing is complete.

    * **Real-Time Mode**: "Original Frame" and "Inference Results" will be displayed simultaneously on the left and right sides.

7.  **Save Results**: After inference is complete, click "File" -> "Save Inference Results."



---



## βš™οΈ Configuring Custom Models



The power of this tool lies in its ability to easily load your own ONNX models.



1.  **Create Directory**: In the root directory where the `.exe` file or `main.py` is located, create a folder named `custom_models`.

2.  **Place Files**:

    * Place your `.onnx` model file (e.g., `my_model.onnx`) into the `custom_models` folder.

    * Create a configuration file named `custom_models_config.yaml` in this folder.

3.  **Edit `custom_models_config.yaml`**:

    The application will automatically create a template (`.template`) of this file upon launch. You can refer to this template for editing. The format is as follows:



    ```yaml

    # ==============================================================================

    # Custom ONNX Model Configuration Instructions (Template Content)

    # ... (Instruction Text) ...

    # ------------------------------------------------------------------------------

    # Example Configuration:

    # ==============================================================================



    custom_models:

      - model_file: "my_custom_model_v1.onnx"  # Ensure this onnx file is in the custom_models directory

        menu_display_name: "My Custom Model - V1 Car Pedestrian"

        class_names: ["car", "person", "truck", "bus"]

        colors: ["#FF0000", "#00FF00", "#0000FF", "#FFFF00"] # Corresponding colors for car, person, truck, bus



      - model_file: "another_custom_detector.onnx" # Another custom model

        menu_display_name: "Another Detector - Object Recognition"

        class_names: ["bottle", "cup", "keyboard", "mouse", "laptop", "monitor"]

        colors:

          - "#E74C3C" # Red

          - "#2ECC71" # Green

          - "#3498DB" # Blue

          - "#F1C40F" # Yellow

          - "#9B59B6" # Purple

          - "#1ABC9C" # Teal

          # ... More colors

    # ==============================================================================

    ```



4.  **Load Models**:

    * Start the application.

    * Click "External Model Management" -> "Load Custom Models."

    * The program will read `custom_models_config.yaml`, verify the existence of the `.onnx` files, and add all valid models to the "External Model Selection" dropdown menu.



---



## πŸ“„ License



This project is licensed under the [GPLv3 License](LICENSE).



## πŸ™ Acknowledgments



* [YOLOv10](https://github.com/THU-MIG/YOLOv10)

* [PyQt6](https://www.riverbankcomputing.com/software/pyqt/)

* [PyQt-Fluent-Widgets](https://github.com/zhiyiYo/PyQt-Fluent-Widgets)

* [ONNXRuntime](https://github.com/microsoft/onnxruntime)