Download Model

May 28, 2026 · View on GitHub

QQ group: 957577822 (full)
QQ group: 546943464 (full)
QQ group: 258137361

I am a Bilibili Up namded kaylordut, you can find me via link

Download Model

There is a repository with github action to converting model, you can find om(huaiwei), rknn(rockchip) and onnx in the repository.

Install Dependencies

PS: The project only supports Ubuntu 22.04. If you wish to support other systems, please send an email to kaylor.chen@qq.com. Consultation fees are required. Thank you.

Add My software source

  • If your device is Jetson(Orin), RK3588 or AiPro
cat << 'EOF' | sudo tee /etc/apt/sources.list.d/kaylordut.list 
deb [arch=arm64 signed-by=/etc/apt/keyrings/kaylor-keyring.gpg] http://apt.kaylordut.cn/kaylordut/ kaylordut main
EOF
sudo mkdir /etc/apt/keyrings -pv
sudo wget -O /etc/apt/keyrings/kaylor-keyring.gpg http://apt.kaylordut.cn/kaylor-keyring.gpg
  • if your device is PC
cat << 'EOF' | sudo tee /etc/apt/sources.list.d/kaylordut.list 
deb [signed-by=/etc/apt/keyrings/kaylor-keyring.gpg] http://apt.kaylordut.cn/kaylordut/ kaylordut main
EOF
sudo mkdir /etc/apt/keyrings -pv
sudo wget -O /etc/apt/keyrings/kaylor-keyring.gpg http://apt.kaylordut.cn/kaylor-keyring.gpg

install software packages

install common packages

sudo apt update
sudo apt install kaylordut-dev libbytetrack libopencv-dev libyaml-cpp-dev

kaylordut-dev: my private Log library based on spdlog.
libbytetrack: ByteTrack library was built by me. you can find it in repository

install ai-instance

  • print ai-instance version
❯ apt policy ai-instance 
ai-instance:
  Installed: 1.0.0-51-gce55bc3-tensorrt
  Candidate: 1.0.0-51-gce55bc3-tensorrt
  Version table:
 *** 1.0.0-51-gce55bc3-tensorrt 500
        500 http://apt.kaylordut.cn/kaylordut kaylordut/main amd64 Packages
        100 /var/lib/dpkg/status
     1.0.0-51-gce55bc3-onnx 500
        500 http://apt.kaylordut.cn/kaylordut kaylordut/main amd64 Packages
  • select ai-instance version
apt install -y ai-instance=1.0.0-51-gce55bc3-tensorrt # if your device is jetson or RTX device, select tensorrt version
# OR
apt install -y ai-instance=1.0.0-51-gce55bc3-onnx # if your device is PC without GPU, select onnx version
# OR
apt install -y ai-instance=xxxxx-rknn # if your device is Rockchip, select rknn version
# OR
apt install -y ai-instance=xxxxx-nnrt # if your device is AI Pro, select nnrt version

please install libnvinfer-plugin10 and libnvinfer10 before installing ai-instance(tensorrt version)
Jetson(Orin) device: https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/arm64/cuda-keyring_1.1-1_all.deb
RTX device: https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/x86_64/cuda-keyring_1.1-1_all.deb

install CUDA Toolkit

Install the CUDA toolkit matching your driver's CUDA version. Use the cuda-toolkit-X-Y meta-package which pulls in compiler, libraries, and tools all at once.

Step 1: check your CUDA version

nvidia-smi | grep "CUDA Version"
# Example output: CUDA Version: 13.0

Step 2: find the matching toolkit package

apt list -a cuda-toolkit 2>/dev/null | grep "cuda-toolkit-"

Step 3: install (replace 13-0 with your version)

# Example for CUDA 13.0:
sudo apt install -y cuda-toolkit-13-0

Version reference (current as of writing):

  • CUDA 12.6 → cuda-toolkit-12-6
  • CUDA 12.9 → cuda-toolkit-12-9
  • CUDA 13.0 → cuda-toolkit-13-0
  • CUDA 13.2 → cuda-toolkit-13-2

The major.minor version must match, e.g. CUDA 13.0 → cuda-toolkit-13-0.

install libnvinfer (TensorRT runtime + dev)

Before installing ai-instance (tensorrt version), you need the following packages:

PackagePurpose
libnvinfer10TensorRT runtime library
libnvinfer-plugin10TensorRT plugin runtime
libnvinfer-devTensorRT development library
libnvinfer-plugin-devTensorRT plugin development library
libnvinfer-headers-devTensorRT development headers
libnvinfer-headers-plugin-devTensorRT plugin development headers

The version MUST match your CUDA version — the package version string includes the CUDA suffix (e.g. cuda13.0).

Step 1: check your CUDA version

nvcc --version 2>/dev/null || nvidia-smi | grep "CUDA Version"
# Example output: CUDA Version: 13.0

Step 2: find matching nvinfer packages

apt list -a libnvinfer10 2>/dev/null | grep "cuda$(nvcc --version | grep -oP 'Cuda compilation tools, release \K[0-9]+\.[0-9]+' || nvidia-smi | grep -oP 'CUDA Version: \K[0-9]+\.[0-9]+')"

Step 3: install (replace the version with yours, all 6 packages share the same version)

# Example for CUDA 13.0:
sudo apt install -y \
  libnvinfer10=10.14.1.48-1+cuda13.0 \
  libnvinfer-plugin10=10.14.1.48-1+cuda13.0 \
  libnvinfer-dev=10.14.1.48-1+cuda13.0 \
  libnvinfer-plugin-dev=10.14.1.48-1+cuda13.0 \
  libnvinfer-headers-dev=10.14.1.48-1+cuda13.0 \
  libnvinfer-headers-plugin-dev=10.14.1.48-1+cuda13.0

# After installation, hold the packages to prevent accidental upgrade:
sudo apt-mark hold \
  libnvinfer10 libnvinfer-plugin10 \
  libnvinfer-dev libnvinfer-plugin-dev \
  libnvinfer-headers-dev libnvinfer-headers-plugin-dev

Version reference (current as of writing):

  • CUDA 13.0 → all 6 packages at 10.14.1.48-1+cuda13.0
  • CUDA 13.2 → all 6 packages at 10.16.1.11-1+cuda13.2

Run apt list -a libnvinfer10 2>/dev/null | grep libnvinfer10 to see all available versions for your system.

TEST

mkdir -pv build
cd build
cmake ..

Please modify your configuration in the config directory, such as the model path, camera index, etc.

Test Yolo

Enter the bulild directory, and run the following command

make yolo_mutilthreading_demo -j$(nproc) # compile yolo demo only
./yolo_mutilthreading_demo

Test Depth Anything

./depth_demo