 |
 |
 |
 |
| Bounding Boxes | Segmentation | Pose Keypoints | Oriented Bounding Boxes |
UltraDarkFusion is an advanced GUI for object detection, segmentation, keypoints and computer vision, designed to streamline dataset management, annotation, model training, and real-time inference.
- ✅ Full Segmentation Support – annotate, train, and run inference on masks with YOLO or SAM.
- ✅ Pose Estimation – automatic keypoint detection, annotation, and training-ready exports.
- ✅ Oriented Bounding Boxes (OBB) – draw, label, and auto-generate rotated boxes via YOLO OBB or SAM.
- ✅ Dataset Manager – organize, deduplicate, and augment your datasets with built-in tools.
- ⚡ TensorRT Acceleration – run inference using
.enginemodels for max speed. - 🧠 Multi-Backend Tracking – BoT-SORT and ByteTrack support with
.pt,.onnx, and.engineformats. - 🗣️ Voice-Activated Class Switching – change classes during labeling by saying the name.
- 🌍 Multilingual UI Translation – dynamically switch languages across the entire interface.
- 🎨 Label Anything – supports boxes, segmentation, keypoints, and OBB — all in one project.
- 🎞️ Video & YouTube Processing – auto-label or extract frames directly from video files or YouTube links.
- 🤖 Automatic Annotation – batch-label folders.
- 🧩 Flexible Inference – switch between PyTorch, ONNX, and TensorRT seamlessly.
- 🧼 Cleaner UI – toggle dark/light mode, filter by class, generate heatmaps, and apply custom QSS themes.
Open Command Prompt and run:
winget install git.git (must restart shell after wards)
winget install Kitware.CMake
winget install --id Microsoft.VisualStudio.2019.BuildTools --source winget --override "--add Microsoft.VisualStudio.Component.VC.Tools.x86.x64 --quiet --wait"cd C:\
git clone https://github.com/lordofkillz/DarkFusion.git
cd DarkFusion
git clone --branch 4.9.0 https://github.com/opencv/opencv.git opencv-4.9.0
git clone --branch 4.9.0 https://github.com/opencv/opencv_contrib.git opencv_contrib-4.9.0
mkdir build install- [Download SAM Weights and Files]https://drive.google.com/file/d/1I6VYRf_Dco_A5REtTBkvjizFTwHmcaeM/view?usp=sharing
- Extract to
C:\DarkFusion\UltraDarkFusion
- Extract to
- Download and install Anaconda.
- Accept default installation settings.
- for experienced devs, just create and environment and pip install -r requirements.txt
- Run
fusion_install.batas administrator to create and configure the environment:
cd C:\DarkFusion\
fusion_install.batCompile OpenCV with CUDA (Optional) **do not procced past this line if you do not know how to build opencv dnn or trouble shoot problem this is depreciated and will be removed in later version. **
To compile OpenCV with CUDA acceleration (recommended for improved performance when using .weights):
- Install CUDA 12.8
- Download cuDNN compatible with CUDA 12.8 from NVIDIA Developer
- Extract and copy cuDNN files to:
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.8
download opencv and opencv-contrib versions with the same as your python versions place both both folders in c:\ create a build and install folder
cd C:\DarkFusion\
fusion_cuda.bat- Install CUDA 12.8
- Download cuDNN compatible with CUDA 12.8 from NVIDIA Developer
- Extract and copy cuDNN files to:
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.8
- Install Darknet by Hank-AI.
- UltraDarkFusion Discord: Join Discord
Special thanks to these open-source projects:
Inspired by DarkMark, developed by Stephane from Hank-AI.
- Remove OpenCV DNN and create a Python wrapper for Darknet
- Remove the document-based help and train a chatbot for assistance
- Improve code efficiency (Ongoing...)
- Upload UltraDarkFusion to PyPI for easier installation and distribution