build.md

Build From Source (Advanced Users)

If you want to compile the project yourself or modify code, follow these instructions.

1. Requirements

• Visual Studio 2026 Community (Download)

• Windows 10 or 11 (x64)

• Windows SDK 10.0.26100.0 or newer

• CMake (Download)

• OpenCV 4.13.0

• [For CUDA version]

  • CUDA Toolkit 13.1
  • cuDNN 9.17.1
  • TensorRT-10.14.1.48

• [For DML version]

  • Use prebuilt OpenCV package (include/lib/dll), no OpenCV build required.
  • One-time setup script: powershell -ExecutionPolicy Bypass -File tools/setup_opencv_dml.ps1

• Other dependencies:

  • simpleIni
  • serial
  • GLFW
  • ImGui

2. Placement of Third-Party Modules and Libraries

Before building the project, download and place all third-party dependencies in the following directories inside your project structure:

Required folders inside your repository:

sunone_aimbot_2/
└── sunone_aimbot_2/
    └── modules/

Place each dependency as follows:

Library	Path
SimpleIni	sunone_aimbot_2/sunone_aimbot_2/modules/SimpleIni.h
serial	sunone_aimbot_2/sunone_aimbot_2/modules/serial/
TensorRT	sunone_aimbot_2/sunone_aimbot_2/modules/TensorRT-10.14.1.48/
GLFW	sunone_aimbot_2/sunone_aimbot_2/modules/glfw-3.4.bin.WIN64/
OpenCV	sunone_aimbot_2/sunone_aimbot_2/modules/opencv/

• SimpleIni: Download SimpleIni.h Place in modules/.

• serial: Download the serial library (whole folder). For CMake build, this library is compiled from sources automatically (no separate serial.sln build step).

• TensorRT: Download TensorRT-10.14.1.48 Place the folder as shown above.

• GLFW: Download GLFW Windows binaries Place the folder as shown above.

• OpenCV: For DML, run tools/setup_opencv_dml.ps1 (downloads prebuilt OpenCV package). For CUDA, use your custom OpenCV build with CUDA support.

Example structure after setup:

sunone_aimbot_2/
└── sunone_aimbot_2/
	└── modules/
		├── SimpleIni.h
        ├── serial/
        ├── TensorRT-10.14.1.48/
        ├── glfw-3.4.bin.WIN64/
        └── opencv/

3. How to Build OpenCV 4.13.0 with CUDA Support (For CUDA Version Only)

This section is only required if you want to use the CUDA (TensorRT) version and need OpenCV with CUDA support. For DML build, skip this step — you can use the pre-built OpenCV DLL.

Fast path (recommended):

Use helper script from repository root:

powershell -ExecutionPolicy Bypass -File tools/build_opencv_cuda.ps1 -AutoDetectCudaArch

The script requires CMake generator Visual Studio 18 2026.

Useful options:

• Set architecture explicitly:

  powershell -ExecutionPolicy Bypass -File tools/build_opencv_cuda.ps1 -CudaArchBin 8.6

• Build OpenCV for popular consumer NVIDIA architectures (GTX 16 / RTX 20 / 30 / 40 / 50):

  powershell -ExecutionPolicy Bypass -File tools/build_opencv_cuda.ps1 -CudaArchBin all

• Reuse already downloaded sources:

  powershell -ExecutionPolicy Bypass -File tools/build_opencv_cuda.ps1 -SkipDownload

• Configure only (without build):

  powershell -ExecutionPolicy Bypass -File tools/build_opencv_cuda.ps1 -ConfigureOnly

Manual path:

1. Download Sources

   • OpenCV 4.13.0
   • OpenCV Contrib 4.13.0
   • CMake
   • CUDA Toolkit 13.1
   • cuDNN 9.17.1

2. Prepare Directories

   • Create: sunone_aimbot_2/sunone_aimbot_2/modules/opencv/ sunone_aimbot_2/sunone_aimbot_2/modules/opencv/build
   • Extract opencv-4.13.0 into sunone_aimbot_2/sunone_aimbot_2/modules/opencv/opencv-4.13.0
   • Extract opencv_contrib-4.13.0 into sunone_aimbot_2/sunone_aimbot_2/modules/opencv/opencv_contrib-4.13.0
   • install cuDNN Default install path C:/Program Files/NVIDIA/CUDNN/v9.17

3. Configure with CMake

   • Open CMake GUI
   • Source code: sunone_aimbot_2/sunone_aimbot_2/modules/opencv/opencv-4.13.0
   • Build directory: sunone_aimbot_2/sunone_aimbot_2/modules/opencv/build
   • Click Configure (Choose "Visual Studio 18 2026", x64)

4. Enable CUDA Options

   • After first configure, set the following:

     • WITH_CUDA = ON
     • WITH_CUBLAS = ON
     • ENABLE_FAST_MATH = ON
     • CUDA_FAST_MATH = ON
     • WITH_CUDNN = ON
     • CUDNN_LIBRARY = .../sunone_aimbot_2/sunone_aimbot_2/modules/cudnn/lib/x64/cudnn.lib
     • CUDNN_INCLUDE_DIR = .../sunone_aimbot_2/sunone_aimbot_2/modules/cudnn/include
     • CUDA_ARCH_BIN = See CUDA Wikipedia for your GPU. Example for RTX 3080-Ti: 8.6
     • OPENCV_DNN_CUDA = ON
     • OPENCV_EXTRA_MODULES_PATH = .../sunone_aimbot_2/sunone_aimbot_2/modules/opencv/opencv_contrib-4.13.0/modules
     • BUILD_opencv_world = ON

   • Uncheck:

     • WITH_NVCUVENC
     • WITH_NVCUVID

   • Click Configure again (make sure nothing is reset)

   • Click Generate

5. Build in Visual Studio

   • Open sunone_aimbot_2/sunone_aimbot_2/modules/opencv/build/OpenCV.sln or click "Open Project" in CMake
   • Set build config: x64 | Release
   • Build ALL_BUILD target (can take up to 2 hours)
   • Then build INSTALL target

6. Copy Resulting DLLs

   • DLLs: sunone_aimbot_2/sunone_aimbot_2/modules/opencv/build/install/x64/vc*/bin/
   • LIBs: sunone_aimbot_2/sunone_aimbot_2/modules/opencv/build/install/x64/vc*/lib/
   • Includes: sunone_aimbot_2/sunone_aimbot_2/modules/opencv/build/install/include/opencv2
   • Copy needed DLLs (opencv_world4130.dll, etc.) next to your project’s executable.

4. Notes on OpenCV for CUDA/DML

• For CUDA build (TensorRT backend):

  • You must build OpenCV with CUDA support (see the guide above).
  • Place all built DLLs (e.g., opencv_world4130.dll) next to your executable or in the modules folder.

• For DML build (DirectML backend):

  • Use prebuilt OpenCV package (include/lib/dll) if you only plan to use DirectML.
  • OpenCV runtime for DML must not import CUDA (cudnn/cublas/npp) or GStreamer (gst*) DLLs.
  • If you want to use both CUDA and DML modes in the same executable, you should always use your custom OpenCV build with CUDA enabled (it will work for both modes).

• Note: If you run the CUDA backend with non-CUDA OpenCV DLLs, the program will not work and may crash due to missing symbols.

5. Configure and Build Project

After sections 2-4 are complete, configure one backend with CMake.

Use separate build directories for each backend:

• DML (DirectML):

  powershell -ExecutionPolicy Bypass -File tools/setup_opencv_dml.ps1
  Remove-Item build/dml -Recurse -Force -ErrorAction SilentlyContinue
  cmake -S . -B build/dml -G "Visual Studio 18 2026" -A x64 -DAIMBOT_USE_CUDA=OFF
  cmake --build build/dml --config Release

  For DML build, OpenCV must be built without CUDA and without GStreamer. The setup script downloads prebuilt OpenCV and places it in sunone_aimbot_2/modules/opencv/prebuilt/opencv/build (auto-detected by CMake).

• CUDA (TensorRT):

  cmake -S . -B build/cuda -G "Visual Studio 18 2026" -A x64 `
    -DAIMBOT_USE_CUDA=ON `
    -DCMAKE_CUDA_FLAGS="--allow-unsupported-compiler" `
    -DCUDA_NVCC_FLAGS="--allow-unsupported-compiler"
  cmake --build build/cuda --config Release

Only Visual Studio 18 2026 is supported for this project.

If your dependencies are stored in non-default paths, pass CMake cache variables, for example:

cmake -S . -B build/dml -G "Visual Studio 18 2026" -A x64 `
  -DAIMBOT_OPENCV_INCLUDE_DIR="C:/opencv/include" `
  -DAIMBOT_OPENCV_LIBRARY="C:/opencv/lib/opencv_world4130.lib" `
  -DAIMBOT_ONNXRUNTIME_DIR="C:/packages/Microsoft.ML.OnnxRuntime.DirectML.1.22.0" `
  -DAIMBOT_CPPWINRT_INCLUDE_DIR="C:/Program Files (x86)/Windows Kits/10/Include/10.0.26100.0/cppwinrt"

You can open the generated solution from the build folder (build/dml or build/cuda) if you prefer building from Visual Studio UI.

Run ai.exe from <build-dir>/Release/.

6. Exporting AI Models

• Convert PyTorch .pt models to ONNX:

  pip install ultralytics -U
  
  # TensorRT
  yolo export model=sunxds_0.8.0.pt format=onnx dynamic=true simplify=true
  
  # DML
  yolo export model=sunxds_0.8.0.pt format=onnx simplify=true

• To convert .onnx to .engine for TensorRT, use the overlay export tab (open overlay with HOME).