triton-client-examples

Example client applications for NVIDIA Triton Inference Server.

Installation (Server Side)

Use the download_model.sh scripts in the application directories, to download the models to a server.

download_model.sh [download_path]

If not specified download_path, the model will be downloaded to ./model_repository.

Installation (Client Side)

These application are platform agnostic. You don't need GPU in your platform.

1. Clone this repository.

   git clone https://github.com/MACNICA-CLAVIS-NV/triton-client-examples
   

2. Install the modules on which these applications depend.

   python3 -m pip install -r requirements.txt
   

Usage

Tiny YOLO v2

python3 main.py [-h] [--camera CAMERA_ID] [--width CAPTURE_WIDTH] [--height CAPTURE_HEIGHT] [--url SERVER_URL]

Triton Tiny YOLO v2 Demo

optional arguments:
-h, --help            show this help message and exit
--camera CAMERA_ID    Camera ID (Default: 0)
--width CAPTURE_WIDTH
                        Capture Width (Default: 640)
--height CAPTURE_HEIGHT
                        Capture Height (Default: 480)
--url SERVER_URL      Triton Inference Server URL (Default: localhost:8000)

Densenet Classification

python3 main.py [-h] [--camera CAMERA_ID] [--width CAPTURE_WIDTH] [--height CAPTURE_HEIGHT] [--url SERVER_URL]
            [--count CLASS_COUNT]

Triton Tiny YOLO v2 Demo

optional arguments:
-h, --help            show this help message and exit
--camera CAMERA_ID    Camera ID (Default: 0)
--width CAPTURE_WIDTH
                        Capture Width (Default: 640)
--height CAPTURE_HEIGHT
                        Capture Height (Default: 480)
--url SERVER_URL      Triton Inference Server URL (Default: localhost:8000)
--count CLASS_COUNT   Class Count to Display (Default: 3)